Integrated Human Practices - Technical
April
Validating the Problem with Professional Stakeholders
Dr. Steffanie Strathdee
After determining our problem with patients, our next step was to talk with leaders in the industry and those who pioneered phage therapy for the masses. This led us to look no further than our longtime collaborator, Dr. Steffanie Strathdee.
A Canadian-born infectious disease epidemiologist, Dr. Strathdee holds a PhD in Epidemiology from the University of Toronto and is currently the Harold Simon Distinguished Professor at the University of California San Diego School of Medicine. She co-directs the Center for Innovative Phage Applications and Therapeutics (IPATH), the first dedicated phage therapy center in North America. Her influence spans decades of work in HIV prevention among marginalized publications and accolades, including TIME Magazine’s “50 Most Influential People in Health Care” (1). Her most compelling contribution came in 2016, when she orchestrated a groundbreaking international effort to save her husband from a deadly superbug using phage therapy, written in her memoir, ‘The Perfect Predator’. This same book started our president’s interest in phage therapy years ago. With her, we learned about the many challenges we can anticipate and the urgency of the problem we are tackling, described as follows:
Barriers in Phage Screening and Access
Dr. Strathdee emphasized that one of the most pressing challenges in phage therapy is the absence of a centralized phage library. At present, most laboratories maintain their own niche collections, forcing clinicians to reach out individually to multiple labs in search of suitable lytic phages. This fragmented system creates significant delays, particularly when cultures must be shipped across institutions for testing. In urgent cases such as sepsis, where patients may only have 48 hours, these delays often mean that patients die before treatment can be administered.
Compounding the issue, certain bacteria are difficult to culture, especially when patients are already on suppressive antibiotics, making it even harder to match pathogens with effective phages. A streamlined system, potentially based on genomic sequencing rather than culture, could dramatically accelerate the process and improve survival outcomes.
Ethical, Regulatory, and Scientific Complexities
The regulatory landscape further complicates phage therapy. Some countries prohibit the use of genetically modified phages, while others, such as the United States, evaluate synthetic and natural phages under the same framework, focusing primarily on the presence of harmful genes such as antimicrobial resistance or toxins. Dr. Strathdee noted that our designed synthetic phages may, in fact, be safer, as they can be engineered to include only functional genes with known protein products. However, global inconsistencies in GMO policies, as well as concerns about downstream effects on the microbiome, continue to slow down adoption. Despite these challenges, phage cocktails, sometimes combining synthetic and natural phages, are being explored as a promising approach.
Patient Experience and Urgency of Innovation
From the patient perspective, from Dr. Strathdee’s experience, phage therapy is often a last resort. Many individuals seeking treatment have endured years of ineffective or toxic antibiotic regimens, with some requiring daily hospital visits. This desperation underscores the need for more accessible and efficient phage therapy pathways. Dr. Strathdee’s own experience, when her husband’s life was saved by phage therapy after all antibiotics had failed, illustrates both the potential of this treatment and the systemic barriers that prevent timely access. As she emphasized, the key takeaway is to “not take no for an answer” and to think beyond conventional frameworks. For stakeholders, this highlights the urgent need to invest in infrastructure, regulatory clarity, and innovation to make phage therapy a viable and scalable solution.
Reflections or Next Steps
Dr. Strathdee’s insights directly validated our approach. Her emphasis on the lack of a centralized phage library confirmed the need for us to design our dry lab with a centralized phage bank at its core. This ensures that instead of replicating the fragmented collections of today, our work focuses on building the infrastructure clinicians actually need: rapid, coordinated access to phages.
Dr. Greg German
For further advice and input, we followed up with Dr. Greg German to ask similar questions.
Dr. Greg German chairs the Phage Therapy Steering Committee for the Association of Medical Microbiology and Infectious Diseases Canada and co-founded Phage Canada, a national initiative accelerating phage science and public engagement. He holds both an MD and a PhD, is a Fellow of the Royal College of Physicians of Canada, and studied medicine in the UK. As an Associate Professor at the University of Toronto and a practicing infectious disease physician at Unity Health Toronto, Dr. German leads Canada’s only chronic infection and phage therapy clinic at St. Joseph’s Health Centre, which includes missions that align heavily with iGEM Toronto Human Practices (2). His clinical trials, biobank development, and regulatory work are laying the foundation for Canada’s future in personalized phage medicine.
Challenges in Screening and Clinical Trials
Dr. German emphasized that screening phage candidates remains a significant barrier in phage therapy. For example, while only two phages can cover up to 90% of Staphylococcus aureus strains, hundreds of phages may be required to achieve similar coverage for E. coli. Defining sensitivity is also problematic, as there are no clear guidelines on how strongly or for how long a phage must kill bacteria to be considered effective. The process is labour-intensive, and phage interactions with antibiotics complicate treatment since certain antibiotics suppress bacterial machinery that phages rely on to replicate. Clinical trials face additional obstacles, including limited funding, regulatory hurdles, and the absence of a compassionate access pathway in Canada. Unlike the United States, where patients can access phages under emergency use, Canadian patients must go through individualized clinical trial processes, which are time-consuming and resource-heavy.
Patient Perspectives and Public Perception
According to Dr. German, patients who have received phage therapy are often extremely grateful, but many others are frustrated by the lack of access. Some even travel abroad to countries such as Georgia or Poland to receive treatment.
Despite this, public perception continues to present challenges, as phages are sometimes stigmatized as “dirty” or associated with outdated practices from Eastern Europe. Dr. German stressed the importance of education, success stories, and publications to demonstrate that phages are natural, widely present in the environment, and have a long history of safe use. While concerns remain about phages carrying antimicrobial resistance or toxin genes, he argued that randomized controlled trials are essential to validate personalized phage therapy and build broader acceptance.
Future Directions and Opportunities
Looking ahead, Dr. German predicted that phage therapy will evolve toward easier production models, improved diagnostics, and the use of artificial intelligence to accelerate screening and treatment design. He noted that engineered phages may eventually expand host range or even redirect one bacterial species against another. Canada, despite its regulatory delays, has played a leadership role through initiatives such as Phage Canada, which grew during the COVID pandemic and continues to engage both researchers and the public. Dr. German concluded that while the field is ambitious and faces what he called a “phage rush,” the growing urgency of the antibiotic resistance crisis will drive innovation, investment, and eventual mainstream adoption.
Reflections and Next Steps
Dr. German’s critique of Canada’s lack of a compassionate access pathway was pivotal. Without such a mechanism, patients in Canada face long delays, while those in the United States can apply for emergency access through the FDA. His frustration with this gap encouraged us to look at other countries for inspiration. In Belgium, phages are prepared on a case-by-case basis in pharmacies under the magistral preparation model. In Georgia, phage therapy has been a routine part of clinical practice for decades. These examples showed us that there are practical ways to make phage therapy available without waiting for decades of randomized controlled trials.
Building on this, we brought these lessons into our entrepreneurship meetings and realized that compassionate use settings were the most realistic starting point for our work. After further research, we structured our business model to begin within research laboratories supporting compassionate use cases in Canada, and to later expand operations to Georgia and Belgium, where clinical integration pathways and public familiarity with phage therapy are already established. This staged approach allows us to refine our screening and regulatory framework domestically while positioning for international scalability.
May
Precautions Before the Project
Dr. Mariana Piuri
Once we determined the urgency of our project, we understood that diving directly into developing our wet lab, dry lab, and hardware workflows would be premature. Instead, we began by identifying potential challenges through targeted interviews, starting with Dr. Mariana Piuri.
Dr. Mariana Piuri is a distinguished microbiologist and Assistant Professor at the University of Buenos Aires, internationally recognized for her pioneering work in bacteriophage biology and synthetic phage engineering. With over two decades of research experience, she has authored numerous influential publications, including the widely cited development of BRED (Bacteriophage Recombineering of Electroporated DNA), a groundbreaking tool that revolutionized the construction of mutant and recombinant bacteriophage genomes. Her research spans from fundamental studies of phage–host interactions to applied strategies for combating antimicrobial resistance, positioning her as the perfect stakeholder for us to further our understanding of phage therapy.
Limits of RBP-Only Approaches
Dr. Piuri emphasized that altering receptor binding proteins (RBPs) is an important first step but insufficient on its own. While RBPs determine host recognition, successful infection also requires DNA injection, replication, transcription, and translation within the host. She cautioned against assuming that modifying RBPs alone could yield a universal phage, as host compatibility depends on multiple genetic and biochemical factors. Her advice was to treat RBP engineering as a useful but partial strategy, and to remain realistic about its limitations when designing workflows.
Technical Considerations for Phage Construction
On the technical side, Dr. Piuri recommended Gibson assembly over BRED for constructing recombinant phages, noting that Gibson is more efficient since it produces only the recombinant phage. She also advised leveraging E. coli as a model system because of the availability of receptor mutants, which can help test different RBPs. However, she warned that extrapolating results from E. coli to other Gram-negative bacteria may be overly ambitious, since differences in cell wall structures and receptor types (proteins versus sugars) can significantly affect host range predictions. She suggested using bioinformatics tools and curated phage sequence collections to better infer host specificity.
Personalized Therapy and Clinical Realities
Dr. Piuri stressed that phage therapy is most effective when personalized. She pointed to clinical experiences where cocktails of phages, or phages with multiple RBPs, were required to address variability within and between patients. Resistance can emerge even during treatment, and efficacy is influenced by patient-specific factors such as immune response. She highlighted the importance of clean phage preparation, warning that endotoxins like LPS can complicate therapy if not carefully controlled. Her advice was to focus on workflows that anticipate variability, incorporate redundancy (such as cocktails or multiple RBPs), and prioritize rigorous preparation standards.
- Paper suggested: Engineered phage with antibacterial CRISPR–Cas selectively reduce E. coli burden in mice (Gencay et al., 2023)
Reflection or Next Steps
Dr. Piuri encouraged us to remain cautious about overpromising commercialized universal phage cocktails, given the complexity and cost of synthetic phage production. Instead, she advised focusing on personalized approaches, supported by strong bioinformatics pipelines and careful experimental validation. She also underscored the importance of building networks, citing examples from European collaborations that pool strains and expertise. Finally, she reminded the team that phages and bacteria coevolve, meaning that new phages will always be needed to overcome resistance. Integrating this evolutionary perspective into the workflow will help ensure adaptability and long-term relevance.
Dr. Kristin Parent
We reached out to Dr. Kristin Parent for more information.
Dr. Kristin N. Parent is a renowned structural virologist and Professor of Biochemistry and Molecular Biology at Michigan State University, where she also serves as the founding Director of the university’s state-of-the-art Talos Arctica cryo-electron microscopy facility. She is internationally recognized for her groundbreaking research on bacteriophages, viral assembly, and host recognition, combining expertise in protein biochemistry, genetics, and advanced cryo-electron microscopy. Her leadership and innovation have been recognized with prestigious honours, including the American Society for Microbiology Young Investigator Award and the JK Billman, Jr., MD Endowed Research Professorship (4).
Regulatory and Economic Barriers
Dr. Parent highlighted that one of the greatest obstacles to widespread adoption of phage therapy is regulatory complexity, particularly in the United States. The FDA requires that every phage be fully sequenced and that the function of each gene be known, which is not feasible for many phages. Even minor genetic changes, such as a single base pair mutation, are considered to create a “new” phage, complicating approval and patenting. She also noted that the for-profit structure of the U.S. healthcare system discourages investment in phage therapy, since it is more costly to identify, tailor, produce, and distribute phages at scale than to develop new antibiotics. Canada, meanwhile, imposes restrictions on genetically modified organisms, further limiting innovation. Dr. Parent advised us to remain mindful of these systemic barriers when framing the potential impact of their work.
Clinical Trials and Combination Therapies
Dr. Parent explained that clinical trials for phage therapy face unique challenges. Placebo-controlled studies may be feasible for mild infections, but they raise ethical concerns in life-threatening cases. Moreover, it is often difficult to determine whether recovery is due to phages, antibiotics, or their combination. She pointed to studies by Paul Turner at Yale, where phages and antibiotics worked synergistically, forcing bacteria into vulnerabilities that allowed one treatment to succeed when the other failed. Her advice was to consider the potential of phage–antibiotic combinations in our workflow, while also recognizing that not all combinations will be beneficial.
Resistance and Evolutionary Dynamics
Dr. Parent confirmed that bacterial resistance to phages is a real and rapid phenomenon, sometimes occurring within 24 hours. Mutations in genes coding for LPS synthesis or surface proteins can quickly block phage binding. While phages can also evolve to overcome resistance, this creates regulatory complications, since the FDA would no longer consider the evolved phage the same therapy as mentioned previously. She advised the team to anticipate resistance in their models and to design workflows that account for evolutionary dynamics rather than assuming static interactions.
Technical Considerations for Modelling and Structural Studies
When asked about modelling approaches, Dr. Parent cautioned that excluding LPS interactions could limit predictive accuracy, since some phages bind specifically to LPS structures. She suggested that protein-protein models may be more reliable, but recommended in vitro studies that knock out LPS synthesis genes to clarify binding mechanisms. In structural studies, she noted that cryo-EM is faster than crystallography for characterizing RBPs, though flexible tail spikes can complicate imaging. Her advice was to carefully select RBPs that are well-behaved in the lab and to validate computational predictions experimentally.
Future of Phage Therapy and AI Integration
Looking ahead, Dr. Parent expressed cautious optimism. She believes phage therapy will remain a last-resort treatment but will grow in importance as antibiotic resistance worsens. She emphasized that high-throughput methods and standardized pipelines will be crucial for scaling phage therapy. Regarding AI, she advised using it as a complementary tool: AI can narrow down candidate RBPs, but wet lab validation remains essential. She also encouraged us to start with classical, well-studied RBPs to benchmark their AI models against established research.
Public Perception and Strategic Advice
Dr. Parent noted that healthcare companies are likely to understand the value of AI-driven phage research more readily than the general public. She observed that older generations may be more skeptical of new technologies, while younger generations are more open to them. Her advice was to tailor communication strategies accordingly. She additionally encouraged us to consider agricultural applications.
Reflection or Next Steps
Dr. Parent confirmed that bacterial resistance to phages can arise extremely quickly, sometimes within a single day, through mutations that alter surface proteins or LPS structures. This motivated us to find ways to enable our model to generate protein receptors that are harder for a bacteria to evolve against. One of the ways to do this is to integrate functionalities into the model to explore the probabilistic landscapes of possible RBP configurations to find the most optimal RBP that binds to bacteria receptors the tightest. This validates our decision to use MCMC as this allowed our model to have more coverage of high-performing RBP types.
June
Phage Therapy
Understanding Phage-Infection Interactions
At this point, we began our wet lab and dry lab workflows.
Roughly two weeks in, the issue we were tackling was the risk of oversimplifying phage–host interactions by focusing too narrowly on receptor binding proteins. Our workflow, centered on AI-driven predictions of RBP–host specificity, may have been vulnerable to producing results that looked promising computationally but failed in vitro because they did not account for the following possible downstream barriers.
Dr. Graham F. Hatfull
So to further our understanding of the phage therapy landscape, we consulted with Dr. Graham F. Hatfull, one of the world’s foremost authorities on bacteriophages. He is the Eberly Family Professor of Biotechnology at the University of Pittsburgh and an Investigator with the Howard Hughes Medical Institute, where he has led groundbreaking research on mycobacteriophages: viruses that infect mycobacteria, including the pathogen responsible for tuberculosis. Recognized with numerous honours, including election to the National Academy of Sciences and the American Academy of Arts and Sciences; we knew he was the perfect consultant for our iGEM project.
Q1. Are phage infection profiles consistent across different bacterial hosts?
Dr. Hatfull explained that phage infection profiles are highly variable and depend strongly on the bacterial pathogen or host in question. In some cases, specificity is largely determined by receptor availability and receptor binding proteins (RBPs). However, in many other cases, receptor recognition is only the first step, and additional bacterial defence systems come into play. While restriction-modification (RM) and CRISPR-Cas systems are the most widely known, he emphasized that bacteria possess a wide array of defence mechanisms that act after receptor binding. These include abortive infection systems, toxin-antitoxin modules, and other poorly understood host defence pathways. He advised us not to oversimplify phage–host interactions as being determined solely by RBPs, but instead to integrate a broader view of bacterial defence systems into our predictive models.
Q2. What are the main bottlenecks in applying phage therapy to different bacterial systems?
Dr. Hatfull highlighted that bottlenecks differ depending on the bacterial system. For non-tuberculous mycobacteria (NTMs), the key limitation is the scarcity of therapeutically useful phages and their narrow host ranges across clinical isolates. This makes it difficult to find broadly effective phages. In contrast, for Mycobacterium tuberculosis (TB), the challenge is not the lack of phages but rather the difficulty of establishing and funding clinical trials. He advised us to tailor our workflow to the specific bacterial system under study, recognizing that the obstacles are not uniform across pathogens.
Q3. How useful are AI-based predictions of phage specificity?
According to Dr. Hatfull, the utility of AI-based predictions varies enormously depending on the bacterial–phage system and the degree of bacterial variation. For hosts with limited genetic variability, such as M. tuberculosis, AI predictions may work well but are not strictly necessary, since phage–host interactions are more predictable. For NTMs, however, where genetic variability is much greater, AI-based systems could be highly valuable. He cautioned, though, that the effectiveness of AI will depend on the size and quality of the training datasets. Current databases are likely too limited to capture the full complexity of phage infection determinants. He advised us to use AI as a complementary tool, but to remain cautious about overinterpreting predictions until larger, more representative datasets are available.
Q4. Is receptor binding sufficient for productive phage infection?
Dr. Hatfull stressed that receptor binding alone is not sufficient for productive infection. While RBPs are critical for initial recognition, successful infection requires multiple downstream processes, including DNA injection, replication, transcription, and translation. He noted that bacterial defences can block infection at any of these stages. However, he acknowledged that in some cases, swapping out RBPs can alter or even expand host range, making RBP engineering a promising but partial strategy. He advised us to integrate RBP-focused approaches with a broader understanding of post-binding infection dynamics, ensuring that our models and workflows do not assume binding equals infection.
Q5. What advice would you give for integrating these insights into our workflow?
Dr. Hatfull encouraged us to design workflows that account for variability across bacterial systems rather than seeking a one-size-fits-all solution. He advised us to prioritize pathogens where AI predictions are most likely to add value, such as NTMs with high genetic diversity. He also recommended that we build larger, high-quality phage–host interaction datasets to strengthen AI models, while always validating predictions experimentally. On the technical side, he suggested exploring RBP swapping as a way to expand host range, but cautioned that this should be paired with careful study of downstream infection processes. Finally, he emphasized the importance of aligning our research goals with the specific bottlenecks of each bacterial system, whether that is phage scarcity, regulatory hurdles, or clinical trial design.
Reflection or Next Steps
Based on the feedback we received from experts, we made several key decisions to strengthen our project. This, in part, motivated our generative page bank idea. Instead of trying to generate RBPs de novo, utilizing the phage bank approach before generation pre-selects phages from nature with desirable therapeutic properties (ex: lytic, no toxin genes, etc.). Thus, the phage we swap our RBPs into is more amenable to treating the patient, allowing our model to make considerations beyond only generating new RBPs.
Artificial Intelligence
Learning about Markov Chain Monte Carlo (MCMC)
Michael (Jun Sub) Lee (Advisor)
At this point, we were tackling the use and implementation of MCMC. Our MCMC setup allowed variable sequence lengths and de novo exploration too far from native RBDs, risking outputs with little biological relevance or proper folding/binding. To answer our questions, we consulted Michael (Jun Sub) Lee, an advisor from the Kim Lab, who greatly helped us with MCMC usage, including how to fix errors, improve the model, address low data issues, and when to make the decision of fine-tuning existing models or building new ones. We discussed the following questions with him across multiple meetings, which occurred from June to September:
Why is RF Diffusion not a suitable generative model for novel RBP structures?
At the start of the Mystiphage project, we considered using RFdiffusion to design glycan-targeting RBPs but ran into major limitations. Michael provided key feedback that led us to reconsider this approach. He pointed out that RFdiffusion depends on having bound protein-ligand structures as training/finetuning data, yet there are fewer than ten glycan-bound solved RBP structures available. Fine-tuning unbound proteins would require significant computational resources, be GPU-intensive, and likely result in very low experimental hit rates (around 1–2 out of 200). Michael also highlighted that hallucination-based models like Boltz-1/2 or BindCraft not only achieve higher hit rates (15–20%) but also allow for co-designing sequence and structure, making validation easier and eliminating the need for docking. His advice on using conserved scaffolds, domain-specific constraints, and glycan-aware scoring was crucial in shifting our pipeline toward biologically relevant and experimentally practical designs. This guidance enabled us to focus on rapid in silico sampling and select high-confidence candidates, establishing the core of the PHORAGER workflow.
What MCMC issues might undermine biologically viable RBP design?
Michael cautioned that MCMC sampling can drift into non-biological sequence space if we do not enforce constraints. Allowing variable sequence lengths, longer or shorter than native RBPs, invites misfolding, loss of modular architecture, and failure at critical steps such as genome injection. He advised us to constrain the chain with hard priors on length, domain boundaries, and topology, for example, preserving β-helix or trimeric fibre architecture when relevant. He also suggested temperature tuning and adaptive proposals that bias acceptance toward sequences maintaining native-like secondary structure and binding motifs. In practice, this means fixing or tightly bounding sequence length to the native RBP or domain of interest, enforcing fold class and domain order to penalize sequences that violate baseplate or fibre modularity, using predicted structure metrics such as pLDDT and contact maps to guide acceptance and keep the chain near plausible folds, and monitoring convergence with structural and biophysical scores to halt runs that drift from native-like basins.
Are there biological “tricks” to keep generative designs close to native RBD function?
Michael emphasized that the further we moved from native receptor-binding domains (RBDs), the faster the biological relevance drops. Rather than pursuing de novo design, he recommended leveraging all available RBD sequences to build a multiple sequence alignment (MSA) and derive positional amino acid frequencies. We should lock down conserved residues, motifs, and glycan or protein-contacting surfaces, while allowing diversity in variable loops that modulate specificity. This involved protecting high-conservation positions from mutation, focusing mutations on low-conservation, tolerance-rich positions, mapping known receptor-contact residues and capping mutation rates there, retaining native baseplate or fibre attachment segments and oligomerization interfaces to avoid losing assembly competence, and seeding models from native RBPs or consensus designs so that edits are fine-grained rather than wholesale redesigns.
How should we design MCMC with biological constraints to maximize usable outputs?
Michael advised baking biological priors directly into the MCMC objective, combining sequence statistics such as MSA conservation, predicted structure quality, and receptor-binding energetics. He suggested iterating constraints (tightening or relaxing them) to target a yield where roughly 80% of outputs are in the right configuration, meaning foldable, bindable, and assembly-compatible. This requires composite scoring that integrates sequence terms like KL divergence from the MSA profile to penalize off-distribution residues, structure terms such as confidence and contact-map alignment to the native fold, and binding terms like docking or affinity proxies against the intended receptor. He also recommended running families of chains with different constraint strengths to balance diversity with viability, and post-processing outputs through structure prediction, stability filters, and interface checks before wet lab selection.
What makes an impressive computational pipeline given scarce data and tough targets like glycans?
Michael noted that large DNA foundation models can shine in specific contexts, but our problem is intrinsically data-poor and biologically complex. The impressive part is not training the biggest model but assembling the right tools for this biologically important problem, where robust computational methods do not yet exist. He encouraged us to highlight biology-first design rooted in RBP modularity, MSA conservation, and receptor interface mapping; problem-specific tooling such as glycan-aware scoring, structure-guided mutation, and modular libraries of native RBPs; transparent ablations that clearly compare random versus targeted masking, unconstrained versus constrained MCMC, and ESM3 versus alternatives to show why our choices improve biological plausibility; and end-to-end traceability with versioned data, decisions, and metrics that tie computational outputs to wet lab feasibility.
How do we frame the project with tight timelines for wet lab results?
Michael suggested leaning into firsts that matter, noting that mini-protein design for glycan binding is largely unexplored, and demonstrating a designed protein that binds a defined glycan, even outside the phage context, would be impactful. He recommended staging deliverables, beginning with in silico validation of glycan-binding designs through docking, molecular dynamics stability, and energy scores, followed by expression and binding assays of the protein alone, and finally fusion into the phage context for functional tests. He also advised us to frame a specificity narrative that shows directed specificity toward particular glycans as a novel and valuable milestone, and to benchmark against native RBPs and simple baselines to underscore progress.
Where are pipelines like this headed, and how do we position ours?
Michael sees relevance growing as data scarcity persists, since decades of curated phage–host datasets would be needed to enable truly massive foundation models. Meanwhile, multi-modal methods will emerge, but applied models often underperform in the lab compared to brute-force wet lab screening or evolutionary methods such as phage display. His advice was to adopt a hybrid strategy that pairs constrained generative design with evolutionary methods, using computation to narrow the search and then letting display or selection refine. He also emphasized making realistic claims that highlight marginal model improvements rather than silver bullets, quantifying hit rates against wet lab baselines, and investing in curated, open datasets of RBP-receptor pairs, structures, and functional assays to steadily raise model ceilings.
Is a generative phage bank a good idea, and how should we build it?
Michael endorsed a generative phage bank seeded from native RBPs as a smart extension, far better than random initialization. He encouraged us to seed wisely by starting from diverse, well-characterized RBPs across receptor classes, both protein and glycan. He also recommended modular cataloging of components such as RBPs, baseplates, and fibres with standardized metadata including host range, conserved motifs, structure predictions, binding assays, and safety flags. Finally, he suggested building an open database with interactive maps of sequence space, conservation overlays, and receptor specificity clusters to accelerate community learning and collaboration.
If we can’t get wet lab results, what evidence still makes the case?
Michael advised us to present a rigorous in silico body of work that demonstrates method quality. He suggested ablations and diagnostics that compare random versus targeted masking in training, inference, as well as impact on fold and affinity metrics. He additionally encouraged implementing constraint sweeps that show how MCMC priors improve structural plausibility and interface integrity. He recommended explaining the rationale for model choices such as ESM3 and how they integrate with biology-first constraints. Furthermore, he proposed conducting generalization tests on held-out RBPs and receptor classes while reporting failure modes in addition to how constraints mitigate them. Lastly, Michael advised that we outline pre-registered wet lab validation plans with assays, controls, and milestones to show translational readiness.
What concrete integration steps should we add to our workflow at this point (June)?
Michael recommended that we immediately lock sequence length and fold class in MCMC, enforce domain boundaries and oligomer interfaces, and apply MSA conservation masks to protect critical residues while focusing edits on variable loops. He suggested running constraint ablations to tune for the most viable configurations before wet lab work, seeding a generative phage bank from native RBPs with full metadata and visualization, and planning hybrid validation that combines computational narrowing with phage display or high-throughput screening.
Reflection or Next Steps
We constrained MCMC sampling with biological priors, exploring variables like sequence length and domain boundaries to preserve modular function and prevent drift into non-functional space. We also learned that for the future, we can guide our design by MSAs of native RBDs to locate conserved residues important in binding. Locking these residues will help improve receptor specificity. Scoring integrated sequence conservation, structure confidence, and binding energetics, resulting in the most optimal sequence outputs.
Bridging Computational Generative Models with Biological Relevance in Predictive RBP Design
After our discussion with Michael, the core workflow issue we were grappling with was how to balance diversity versus performance in sequence generation, while ensuring the pipeline remained computationally sound. The choice to use MCMC reflected a desire for broad sequence exploration, but it raised concerns about mixing time (for example, when the chain truly reflects the proposal distribution) and the fact that MCMC often yields diverse but not necessarily high-performing candidates.
Alston Lo and Amardeep Singh
As a result, we turned to Alston Lo, a University of Toronto graduate who excelled in computer science and who is an advisor on the team, as well as Amardeep Singh, a mechanical engineering graduate with a Biomanufacturing emphasis who is currently also a team member.
Pipeline Design Concerns
After introducing our timeline to them, Alston noted that the pipeline design was generally reasonable, but he cautioned that the main caveat with MCMC lies in its mixing time. Without confirming when the chain has converged to the proposal distribution, the generated sequences may not truly reflect the intended sampling space. His advice was to integrate convergence diagnostics into the workflow to ensure the outputs are representative rather than prematurely assuming validity.
MCMC vs. Genetic Algorithms
Amardeep emphasized that MCMC has been validated in prior studies and is a defensible choice for exploratory sequence generation. Alston, however, highlighted the trade‑off: MCMC produces diverse but not necessarily high‑performing samples, while genetic algorithms (especially 3D ones that optimize directly for binding affinity) yield better‑scoring candidates at the cost of diversity. He suggested that the team should explicitly define whether their priority is exploration or optimization, and potentially combine both approaches for balance.
Genes vs. Proteins
Alston strongly recommended focusing on proteins rather than genes. He explained that gene sequences are too long for practical generative modelling, genome‑scale models remain underdeveloped, and proteins are ultimately the functional endpoint of interest. While gene inputs can be reasonable in some contexts, he stressed that for an iGEM‑scale project, protein‑level modelling is far more practical and impactful.
Framing for the Wiki
Alston encouraged framing the project around biological motivations and showing how these informed computational choices. He pointed to decisions such as choosing AF3 vs. Boltz2 for proteins and glycans as examples of biologically grounded reasoning. He stressed that the real innovation was not just running models, but integrating them into a coherent workflow and demonstrating how they could help move toward wet lab applications.
Bridging Computation and Biology
Alston also highlighted the importance of documenting post‑filtering steps and explicitly noting failure modes from a biological perspective. He framed the ultimate vision as a “self‑driving lab,” but acknowledged that every step toward wet lab integration is already frontier research. Including these links in the workflow narrative would underscore the project’s innovation.
Managing Optics Without MCMC
If MCMC was not ready by the wiki deadline, Alston reassured that this would not undermine the project. A working wet lab result would be ideal, but even without it, the team could highlight alternative success criteria and learnings. He emphasized that success should be framed as a continuum of insights, not a binary outcome.
Addressing MCMC Failure Modes
Alston proposed practical fixes for the sequence length collapse issue in ESM3. These included masking out the stop token until a minimum sequence length is reached, or incorporating a length penalty into the reward function. He suggested that documenting this as a failure mode patched by domain expertise would strengthen the project’s credibility and showcase adaptive problem‑solving.
Reflection or Next Steps
We found that the interview validated our decision to prioritize protein-level modelling over gene sequences, reaffirming that focusing on proteins as the functional endpoint was the most practical and biologically meaningful approach for iGEM-scale work.
It also confirmed the soundness of our computational framework, showing that our choices to use Boltz2 were well justified.
July
Designing Evolutionarily Robust Phages
Afterwards, we needed to ensure that our computationally designed phage tails were not only optimized for binding but also evolutionarily robust. The feedback highlighted that our workflow risked oversimplifying phage function by focusing too narrowly on receptor binding, when in reality the tail must also fold correctly, integrate into the phage structure, evade host immunity, and maintain viral fitness. At the same time, we were confronted with the evolutionary dilemma between bacteria and phages: even if our designs worked initially, bacterial populations could rapidly evolve receptor mutations or deploy systems like CRISPR to escape infection.
Dr. Sarah Otto
To address this concern, we consulted with Dr. Sarah Otto, an expert on genetic variation, recombination, and host-parasite dynamics. A MacArthur Fellow and recipient of the Darwin-Wallace Medal, she has built a career at the University of British Columbia (though often collaborating with colleagues at Toronto), developing mathematical and computational models that explain how organisms adapt under constant selective pressure. Her expertise is directly relevant to phage therapy, immunology, and infectious disease because she studies the very principles that govern how bacteria evolve resistance and how viruses, including phages, adapt in response.
Initial Impressions of the Project
Dr. Otto was enthusiastic about the overall concept of our phage therapy project, calling it a great idea. However, she cautioned that our focus on receptor binding alone may be too narrow. A phage tail must do more than simply attach to a bacterial surface, it must also fold correctly, integrate into the viral structure, embed properly into the bacterial cell, and avoid recognition by human antibodies. Her advice was that our machine learning workflow should not only optimize for binding but also test whether engineered phage tails can perform these additional biological tasks.
The Co-Evolutionary Arms Race
As an evolutionary biologist, Dr. Otto emphasized that nothing in the arms race between bacteria and phages is ever perfect. While some phages adopt broad host ranges and perform moderately across many bacteria, our project has the advantage of focusing on a specific phage–bacterium interaction, allowing us to optimize binding with precision. Still, she raised concerns about “gain of function” risks: engineered phages could potentially attach to unintended targets, leading to harmful consequences such as killing beneficial E. coli in the gut. She advised that a major focus of our workflow should be avoiding cross-infectivity between strains in the human body. This means integrating safety checks into our design pipeline to ensure specificity and prevent unintended ecological disruptions.
Matching Alleles and Bacterial Evolution
Dr. Otto drew on her “Matching Alleles” model, which posits that parasites are most successful when they precisely match their host. She explained that while our engineered phages will inevitably drive bacterial evolution, it is not clear whether this will make bacteria “worse.” The real concern is whether bacterial resistance will evolve faster than phage spread. Using her metaphor, if the phage tail is a key that fits the bacterial lock, bacteria can simply change the lock. For our workflow, this means we must anticipate receptor mutations and design phages that remain effective across multiple mutational variants.
Evolutionary Concerns in Therapy Development
When asked whether bacterial evolution would be a major concern in therapy development, Dr. Otto stressed that it must be considered, especially when moving from controlled lab conditions to animal models. Strong selection pressure from effective phages will accelerate bacterial evolution, and given the vast population sizes of bacteria, resistant variants are almost certainly present from the outset. She recommended that our machine learning pipeline be designed to test not only whether a phage tail binds to the current receptor, but also whether it can bind to plausible receptor mutations. By simulating amino acid substitutions one at a time, we could identify phage tails that remain robust across multiple mutational iterations. This would yield designs more resistant to evolutionary escape.
Bacterial Escape Routes
Dr. Otto highlighted several evolutionary escape routes bacteria might use, including CRISPR systems that cut phage genomes and receptor modifications, which are often the fastest and most common resistance strategy. To counter this, she advised that our workflow should explicitly test phage tails against every possible amino acid change in the receptor, ensuring that binding is maintained even as bacteria mutate. This approach would help us design phages that remain effective despite rapid bacterial adaptation.
Speed of Receptor Evolution
She confirmed that receptor changes in bacteria can occur quickly, even during therapy. Bacterial populations already contain significant variation before treatment begins, and because phage therapy cannot reach every bacterium simultaneously, replication and mutation will continue during treatment. This reinforces the need for our workflow to anticipate variation and design phages that can tolerate receptor diversity.
Cross-Reactivity and Safety Concerns
Another evolutionary problem Dr. Otto identified was cross-reactivity. A phage tail designed to bind a bacterial receptor might also bind unintended bacterial receptors or even human receptors. Even if this does not directly harm humans, it would reduce therapeutic efficiency by diverting phages away from their intended bacterial targets. She recommended that our algorithm incorporate negative filters, categorizing what the generated tail should not bind to. Mining protein databases to exclude human and off-target bacterial receptors would be an important safeguard to integrate into our workflow.
Safeguards and Design Considerations
Dr. Otto suggested that we mine large protein databases to train our machine learning models to recognize and avoid non-target receptors. She also emphasized that phage tails must retain their structural integrity and not compromise viral fitness. If engineered tails reduce replication efficiency or stability, the phage will evolve compensatory changes that could undermine our design. To mitigate this, we should design tails that both bind effectively and preserve viral fitness, ensuring that the engineered phage remains competitive in evolutionary terms.
Evolution of Engineered Phages
She warned that even if we engineer a phage tail to bind effectively, it may evolve during therapy. If our design reduces viral fitness, natural selection will favour mutations that restore stability or replication efficiency, potentially undoing our engineering. To prevent this, our workflow should prioritize designs that are as close as possible to the wild-type phage structure while still achieving the desired binding. By aligning engineered designs with evolutionary constraints, we can reduce the likelihood of rapid reversion or destabilization.
Final Evolutionary Considerations
Dr. Otto concluded by stressing the importance of structural fidelity. The wild-type phage is the product of millions of years of evolution, and disrupting its architecture too much could render it ineffective. She recommended that we use tools like AlphaFold to ensure that engineered phage tails remain structurally similar to wild-type while still achieving new binding properties. This balance between innovation and evolutionary realism is critical for ensuring that our designs are both functional and durable.
Integration into Our Workflow
Dr. Otto’s advice pushed us to expand beyond binding optimization and incorporate checks for folding, immune evasion, viral fitness, and cross-reactivity. She emphasized the need to simulate receptor mutations, mine protein databases for off-target risks, and design phages that remain structurally close to wild-type. By integrating these safeguards, our workflow can produce phages that are not only effective in the short term but also resilient to bacterial evolution, safe for therapeutic use, and aligned with the realities of host–pathogen coevolution.
Reflection or Next Steps
This conversation motivated future plans for our team. We have used AlphaFold as an orthogonal validation tool to observe whether our generated RBDs fold properly, but we can also use this tool to confirm that our designed phage RBDs maintain structural similarity to wild-type forms, which would support functional therapeutic performance. We plan to extend this framework to minimize cross-reactivity and off-target effects through negative filtering and database scans to exclude interactions with commensal microbes, which is very relevant to the hardware team’s objective of targeting the gut resistome.
Addressing Public Trust with AI-Generative Innovation
Dr. Ikhimhiagie Felix Askeomhe
At this point in our timeline, we needed insight in how to bridge the gap between computational innovation and clinical trust in new therapies. To help us, we interviewed Dr. Ikhimhiagie Felix Asekomhe, a seasoned family physician with over three decades of clinical experience, practicing in the Greater Toronto Area and affiliated with the University of Toronto’s medical community. He has built his career around comprehensive, patient-centered care that spans prevention, acute treatment, and chronic disease management (7).
The Role of Antibiotics and the Need for New Therapies
Dr. Asekomhe emphasized that while it is important to be prudent with antibiotic use, it is equally critical not to withhold them when they are genuinely needed. He has seen firsthand how patients left untreated for too long can suffer complications, which in turn may worsen resistance. He stressed that science must continue to evolve, producing more potent antibiotics and alternative therapies, because antibiotic discovery has slowed dramatically. His concern is that if current trends continue, we may eventually face complete resistance to antibiotics. For our workflow, this underscored for us the urgency of positioning phage therapy not as a replacement for antibiotics, but as a complementary and future‑proof therapy that addresses the looming resistance crisis.
Patient Perceptions of Phage Therapy
Dr. Asekomhe reflected on how patients might perceive phage therapy, especially in the aftermath of COVID‑19, when viruses are often seen as threats. He acknowledged that every new therapy is difficult to introduce, but if phages prove to be highly effective and free of negative side effects, doctors will adopt them and patients will accept them. Thus, if we ever launch this into a viable startup, we must prioritize generating strong, transparent evidence of safety and efficacy. Building trust will be as important as demonstrating technical innovation, and our communication strategy should anticipate skepticism while highlighting clinical reliability.
Safety, Data, and Clinical Caution
With over 30 years of practice, Dr. Asekomhe has seen many therapies initially hailed as breakthroughs later revealed to be dangerous. His clinical philosophy is to be cautious with new medications, especially those involving bacteria, viruses, and engineered systems. Before prescribing phage therapy, he would want to see robust safety data and long‑term clinical trial results. He also noted that while most patients would be open‑minded, they would need reassurance that the therapy would not have harmful offshoot effects. Importantly, he raised concern about viral evolution and how it might affect safety. This advice translates into embedding evolutionary foresight into our AI pipeline, testing not only for binding efficacy but also for stability, safety, and robustness against unintended evolutionary changes.
AI in Clinical Practice
Dr. Asekomhe expressed openness to the use of AI in medicine, provided it is applied with proper clinical judgment. He believes his patients would also be receptive to AI‑assisted therapies, as long as the doctor remains the decision‑maker. He sees AI as an inevitable and important part of the future of healthcare. For our workflow, this means we should frame our AI‑generative RBP design not as replacing clinicians, but as a decision‑support tool that enhances their ability to deliver safe, effective, and personalized therapies. Transparency in how AI contributes to phage design will be key to building trust among both doctors and patients.
Reflection or Next Steps
In response to this feedback, we incorporated these takeaways by prioritizing transparency and trust-building in all aspects of our communication. Our newsletter and outreach materials were redesigned to clearly convey safety and efficacy data, making complex concepts accessible to clinicians and the public. We framed our AI as a decision-support tool rather than a replacement for clinicians, emphasizing that it strengthens clinical judgment while maintaining human oversight.
August
Ensuring Innovation and Accessibility
At the start of August, we learned that there is a disconnect between laboratory innovation and hospital needs: labs are developing phages for pathogens like MRSA, but hospitals often remain unaware or unable to access these solutions. This meant we had to think beyond just designing effective AI‑generated RBPs and start addressing translation bottlenecks: how to streamline diagnostics, reduce costs, anticipate regulatory and GMP hurdles, and ensure that our computational outputs could be integrated into scalable, clinically relevant delivery systems. In essence, the problem was not only about designing phages that work, but also about embedding our AI pipeline into a healthcare ecosystem where accessibility, affordability, and clinician awareness determine whether patients actually benefit.
Pranav Johri
To help us address this concern, we interviewed Pranav Johri, a pioneering patient‑advocate and entrepreneur whose personal journey with antibiotic‑resistant infection led him to become one of the most visible champions of phage therapy in India. After overcoming a life‑threatening, drug‑resistant condition through treatment at the Eliava Institute, he co‑founded Vitalis Phage Therapy, an organization dedicated to making phage treatment more accessible, affordable, and better integrated into healthcare systems. He understands the science and regulatory hurdles, but also the lived reality of patients navigating resistant infections. During this interview, we also learned about phage therapy in India, solidifying our mission in conducting global outreach. This gave us insight as to how relevant phage therapy is in other countries.
Treatment Timelines and Patient Experience
Reflecting on his own treatment at Eliava, Johri explained that therapy typically begins with a standardized polyphage cocktail, followed by customization if resistant strains are detected. In his case, MRSA and enterococcus responded to the standard cocktail, but S. mitis required a custom monophage preparation. The overall process took about 10 weeks, which is consistent with timelines for other patients. For us, this further highlighted the need to shorten the design-to-treatment pipeline. AI‑driven predictive RBP design could dramatically reduce the time required to generate effective custom phages, making therapy more viable for patients with acute or rapidly progressing infections.
Manufacturing and Delivery Gaps
Johri emphasized that while many labs in India are focused on developing phages, there is a critical gap in treatment delivery. Hospitals often remain unaware of the phages being developed for pathogens like MRSA, creating a disconnect between research and clinical application. His company positions itself as a bridge, linking labs, hospitals, and patients. He also noted that natural phages cannot be patented, which complicates commercialization models, though funding for R&D has improved in recent years. For our workflow, this suggests that integration with clinical delivery systems is as important as innovation in phage design. Our AI pipeline should be framed not just as a discovery tool but as part of a broader ecosystem that connects diagnostics, manufacturing, and clinical deployment.
Regulatory and Infrastructural Challenges
Johri pointed out that regulatory and infrastructural barriers, such as GMP compliance, biohazard classification, and lack of skilled personnel, currently limit the establishment of dedicated phage manufacturing facilities in India. While awareness and funding have improved, the disconnect between lab research and hospital needs persists. For our workflow, this means we must anticipate regulatory hurdles early, ensuring that our AI‑generated phages are designed with manufacturability and compliance in mind, not just theoretical efficacy.
Shifting Awareness and Acceptance
Johri observed that awareness of antimicrobial resistance (AMR) and phage therapy has grown significantly in India over the past decade. Five to seven years ago, physicians were reluctant even to acknowledge AMR as a problem, but COVID‑19 helped break down stigma and accelerated acceptance. Today, hospital management openly acknowledges that a significant proportion of infections, such as urology cases, are untreatable with antibiotics. Internationally, awareness is also rising, driven by the growing prevalence of AMR and the increasing number of publications and case studies on phage therapy. To us, this means that timing is critical, there is now a receptive audience among clinicians and patients, and our AI‑driven approach should be positioned as part of this broader wave of acceptance.
Patient Perceptions and Ethical Concerns
Johri noted that earlier, patients were hesitant about being treated with viruses, but growing awareness of the human microbiome has reduced these concerns. His team now engages with 400–500 doctors annually, and the change in attitudes has been profound. Patients and clinicians are increasingly open to phage therapy, provided it is safe, effective, and accessible. For our workflow, this reinforces the importance of clear communication and transparency. Our AI‑designed phages must be presented not as abstract computational products but as safe, rigorously tested therapies that patients can trust.
Integration into Our Workflow
The central issue we were tackling during this time was how to align our AI‑generative, predictive RBP project with the realities of treatment delivery, accessibility, and patient trust. Johri’s insights highlighted that success in phage therapy is not just about designing effective phages: it is about reducing costs, shortening timelines, bridging the gap between labs and hospitals, and ensuring equitable access across socioeconomic and geographic divides. For our workflow, this means embedding considerations of affordability, manufacturability, and delivery into the design process, while also framing our AI pipeline as a tool that accelerates treatment without losing sight of patient safety and clinical trust.
Accessibility and Cost of Phage Therapy in India
Pranav Johri highlighted that under the current compassionate use model, phage therapy in India remains expensive and largely inaccessible to patients from lower socioeconomic backgrounds. He emphasized that our mission to streamline processes will have a plethora of positive effects. For example, streamlining phage therapy in India will reduce the need for international travel, allow diagnostics domestically, and perform testing in-house. These steps will significantly lower costs and improve accessibility.
Urban vs. Rural Divide in Access
Johri noted that outreach efforts are concentrated in urban centers, particularly in South India, where hospitals and medical institutions have stronger research cultures and greater openness to phage therapy. In contrast, rural hospitals often lack diagnostic infrastructure and tend to refer patients with untreatable infections to urban centers. This highlights a structural inequity: while innovation is concentrated in cities, rural populations remain underserved. For our workflow, this suggests that any AI‑based phage design pipeline must consider deployment models that can function in low-resource settings. For example, rapid diagnostics that can be integrated into smaller hospitals or mobile labs.
Reflections or Next Steps
Pricing Strategy: We determined that an initial price of CAD 3,000 per RBP balances affordability for patients with coverage of manufacturing, regulatory compliance, and delivery costs.
Global Benchmarking: Comparing costs internationally helped us understand what patients and healthcare systems can reasonably pay, informing both our business plan and prioritization of cost-reducing measures such as domestic diagnostics and in-house phage production.
Integrated Accessibility Planning: The pricing decision was coupled with regional hub and mobile lab models to ensure rural and underserved populations could access therapy without prohibitive travel costs.
Integrated Human Practices for Our Shift into a Startup
Disclaimer:
The following section presents interviews and insights gathered from a diverse group of stakeholders, whose expertise informed our understanding of intellectual property, regulatory considerations, technical differentiation, AI strategy, financial projections, market growth, and beyond. While these perspectives were primarily incorporated into the entrepreneurial dimension of our iGEM project, they also played a critical role in shaping the workflows of both our dry lab and wet lab subteams.
Our NEST Entrepeneurship Team
Our Entrepreneurship Team Pitching our Product
Meeting 1:
Linda Drisdelle holds a BASc in Chemical Engineering and an MBA from the Rotman School of Management at the University of Toronto. She currently serves as Chief Operating Officer at Soteria Company, LLC, and previously held the role of General Manager of the Emissions Reduction & Compliance Group at Pinchin. Her expertise lies in hydrogen systems, industrial gases, and environmental compliance, with a strong track record of leadership in sustainability-focused industries.
Milane Segal earned her MBA from the Rotman School of Management at the University of Toronto. She is the co-founder of Synergy Water Solutions, a nonprofit organization dedicated to advancing sustainable water access in Ghana. Her work emphasizes social entrepreneurship, global development, and experiential learning, with a particular focus on creating scalable solutions for underserved communities.
Mark Graham is a Professor at the University of Oxford and the Director of Fairwork. He has contributed as a guest speaker at U of T’s Platforms & Labour Speaker Series. His expertise spans global digital labor markets, the ethics of artificial intelligence, and internet geography, with a strong emphasis on the social and ethical implications of emerging technologies.
Cartik Sharma is a graduate of the Rotman School of Management at the University of Toronto. He is the Chief Technology Officer of Neuromorph and also serves as a mentor at the U of T Hatchery. His expertise covers quantum machine learning, surgical robotics, and medical imaging, with a particular interest in the intersection of advanced computation and healthcare innovation.
Specific Feedback After We Practice Pitched our Project
Feedback from Mark Graham
Mark Graham informed the team that he found the pitch engaging and accessible, even without a deep technical background. He praised the clarity of the storytelling and highlighted the competitive slide as a strong element of the presentation. However, he raised a concern regarding the sustainability of the competitive advantage in generative AI, noting that the technology is becoming increasingly democratized and widely accessible.
Feedback from Milane Segal
Milane Segal expressed that the presentation was strong and effectively conveyed the urgency of the problem being addressed. She acknowledged the clear need for the solution but emphasized the importance of articulating how the product distinguishes itself from competitors. She also questioned whether alternative approaches might allow others to outperform the team, underscoring the need for sharper differentiation.
Feedback from Linda Drisdelle
Linda Drisdelle stated that she was impressed by the overall concept but found the slides visually overwhelming. She recommended simplifying and decluttering the presentation to improve clarity and impact. Additionally, she suggested that the team introduce their credentials earlier in the pitch, as doing so would enhance credibility and establish trust with the audience from the outset.
Feedback from Cartik Sharma
Cartik Sharma praised the clarity of the presentation and commended the team for effectively communicating a complex idea. He raised a forward-looking question regarding regulatory pathways, specifically asking about the team’s plans for FDA approval in the context of AI-driven protein design.
Technical Differentiation and AI Strategy
We explained that most private competitors rely on template-based approaches, whereas Mystiphage focuses on protein-specific binding, which offers greater precision and scalability. They emphasized that their AI model is built on open-source foundations but is designed to outperform existing systems through a custom protein learning architecture. While quantum computing had not initially been considered, we acknowledged its potential integration, particularly for generative and predictive modeling. As a benchmark, they referenced the Boltz 2 model from MIT, which has demonstrated superior binding accuracy.
Competitive Advantage and Intellectual Property
Linda Drisdelle advised the team to file broad patents early in order to secure strategic ground and protect their innovations. Cartik Sharma contributed additional insights by referencing the use of quantum GANs and highlighting local companies such as Protein Cure, which are also working in protein prediction. These examples underscored the importance of maintaining a technological edge.
Reflection or Next Steps
Based on our stakeholder discussions, we incorporated their recommendations to improve both our technical and entrepreneurial strategy. We refined our pitch structure and visuals to communicate complex ideas clearly while emphasizing our team’s credentials to establish credibility. We strengthened our competitive differentiation by highlighting Mystiphage’s protein-specific AI model, benchmarked against the Boltz 2 architecture, and explored future enhancements such as quantum computing for generative and predictive tasks. On the regulatory front, we investigated the FDA’s Special Access Program as a potential pathway for early deployment and aligned our workflow with regulatory considerations. Collectively, these actions reinforced the importance of integrating AI-driven technical design with market positioning, regulatory strategy, and startup readiness, ensuring our project is both scientifically rigorous and practically implementable.
Business Model and Funding
Both Linda Drisdelle and Milane Segal stressed the importance of clarifying the revenue model and articulating a clear monetization strategy. Lisa explained that current funding is supported by the University of Toronto and that the business model is being refined through ongoing conversations with laboratories and phage therapy firms. She emphasized that the therapy should be framed as additive to antibiotics rather than a replacement, as this positioning helps shape demand more effectively. We added that initial funding would be directed toward research and development, as well as navigating regulatory pathways, with the expectation that the company would be well-positioned once regulatory barriers begin to ease.
Meeting 2:
Erika J. Murray serves as the Go-To-Market Lead at the University of Toronto’s Entrepreneurship Hatchery. She is both an intellectual property lawyer and a PhD-trained chemical engineer, with extensive experience in technology startups. She provides strategic commercialization and legal advice to high-growth ventures, teaches intellectual property and business law, and is an active mentor and board member in youth and innovation communities.
Balaji Gopalan is the Co-Founder and Chief Executive Officer of MedStack, a leading privacy compliance platform for digital health. He is an expert in product strategy and platform ecosystems, with prior experience launching BlackBerry’s BBM globally. At the University of Toronto, he contributes as a mentor and speaker at the Health Innovation Hub (H2i) and Hatchery events, and is a strong advocate for healthcare innovation and startup mentorship.
Sharon Barney is the President of Sharon Barney Consultancy, a firm dedicated to advancing gender equity in construction, engineering, and technology. She holds a B.A. (Hons.) in English from the University of Toronto, as well as a B.Com and MBA from the University of Windsor. With over 25 years of experience in the construction industry, her career spans general contracting, building automation, and project financing. She has also worked on the installation of Siemens MRI and CAT scan equipment and continues to consult on development projects.
Joseph Orozco is the Executive Director of Entrepreneurship and the founder of The Entrepreneurship Hatchery at the University of Toronto. He also teaches entrepreneurship at the Faculty of Applied Science & Engineering. He developed the FEEL™ methodology for venture formation and entrepreneurial human capital. With over 30 years of experience as a founder, operator, educator, and angel investor, he has led multiple ventures, including Telequote (sold to Reuters), Latamtrade, and Simeon International.
Pitch Content and Presentation
The mentors observed that too much time was spent explaining basic phage concepts, which have been known for decades. The presentation lacked clarity on Mystiphage’s unique value proposition, particularly when compared to competitors such as MIT and Recursion. The scientific explanation was not sufficiently differentiated from other startups or labs working in phage therapy and drug discovery. The team was advised to focus on articulating “why this, why now”, emphasizing the urgency and distinctiveness of Mystiphage’s approach.
Intellectual Property Strategy and Technical Leadership
The use of open-source models was questioned, with the mentors requesting a clear rationale for this choice and an explanation of how the team intends to contribute back to the community. The intellectual property strategy was described as underdeveloped, with no licensing decisions yet made regarding the model. Critical feedback emphasized that the technical lead must play a stronger role in shaping both the pitch and the IP discussion.
Competitive Landscape
Stakeholders highlighted several comparative companies: Phagos, PhageLab, PhageAI, and B-Phage, which are AI-driven platforms for matching phages with bacteria, as well as Locus Biosciences, Tolka AI, and Intralytix, which employ novel methods such as CRISPR and robotics. It was suggested that Mystiphage strengthen its competitive framing by including a 2x2 matrix comparing its strategy with those of pharmaceutical companies, government labs, and direct competitors.
Business Model and Go-to-Market
The pitch was seen as lacking a clear and simplified business plan that demonstrates logical steps from research to commercialization. Our mentors noted that the revenue timeline may be four to five years away due to FDA and IP hurdles. Key questions were raised about who would pay for phage-matching services and how Mystiphage would avoid competing directly with hospital labs. Suggested go-to-market strategies included pursuing a licensing model or forming channel partnerships with large pharmaceutical or digital health companies.
Science versus Business Framing
Balaji Gopalan emphasized that investors must understand Mystiphage’s scientific edge and intellectual property potential, noting that this is not a typical SaaS model. Sharon Barney urged the team to provide a clearer explanation of the process behind phage–bacteria matching and how resistance is addressed. Leslie Jamieson added that pharmaceutical companies have largely avoided investing in antibiotics due to resistance challenges, and MystiPhage must clearly articulate what it is doing differently to overcome this barrier.
Regulatory Strategy
Stakeholders expressed concern about long regulatory timelines and uncertainty surrounding FDA and Health Canada approval. The recommendation was to be transparent that regulatory approval is years away and to avoid overstating readiness. Instead, the team should emphasize continued alignment with evolving policy, potentially through the involvement of an advisor with relevant regulatory experience.
Competitor Landscape and Market Positioning
Specific competitors were highlighted, including Tolka AI Therapeutics, which develops personalized phage therapies for chronic infections; Locus Biosciences, which uses CRISPR-Cas3 phage design and is in UTI trials; and Intralytix, which combines robotics and bioinformatics for discovery. Stakeholders urged Mystiphage to clarify why major pharmaceutical players such as Pfizer have not yet tackled this problem, what Mystiphage is doing differently, and how resistance represents an urgent, large-scale problem beyond niche diseases.
Business Model, Revenue, and Market
Investors repeatedly asked for missing figures, including market size, projected revenue, and cost breakdowns. Stakeholders suggested including specific market segments, estimated total addressable market (TAM), financial projections, and the cost structure of building and maintaining a phage model. A clear timeline and benchmarks for productization were also recommended.
Intellectual Property Strategy
The team explained that their IP focus is on the proprietary generative model rather than individual phages. We noted that while the foundation is open-source, such as collaborations from MIT and Recursion, Mystiphage is building a unique protein-learning model akin to large language models. Investors raised concerns about whether the IP is sufficiently strong if based on open models, urging the team to clarify ownership, contributions, and safeguards against future disputes over code. Stakeholders stressed the importance of articulating the model’s defensibility and innovation clearly.
Regulatory Landscape
Investors cautioned against regulatory optimism, suggesting that a realistic timeline for FDA or Health Canada approval may be five to seven years or longer. They advised avoiding phrasing such as “will be approved” and instead using language like “we are monitoring policy changes.” Proposed solutions included engaging an advisor with Health Canada experience, indicating a phased alignment strategy (such as early-stage lab interactions), and clarifying whether an Investigational New Drug (IND) application would be required and when it might be pursued.
Science Communication and Clarity
Stakeholders strongly recommended simplifying the pitch for non-biotech investors. They advised illustrating the concept visually, for example, by showing how binding occurs and how the therapeutic application works. The team was encouraged to clearly explain how Mystiphage’s approach to protein generation solves a real-world problem, while avoiding overly complex jargon or assumptions about audience knowledge. Including a business logistics slide to connect the science with the commercial pathway was also suggested.
Reflection or Next Steps
Based on stakeholder feedback, we restructured our pitch and refined our business framing. We simplified technical explanations to focus on Mystiphage’s distinct value proposition and clarified how our protein-specific AI model differs from other competitors in phage therapy. We added a competitive matrix that visually compares our approach with pharmaceutical companies, government laboratories, and direct competitors.
We reframed our intellectual property strategy to emphasize the proprietary nature of our generative protein learning model and outlined safeguards that protect ownership while maintaining collaboration with open source foundations
Meeting 3:
James Macintosh is the former Canadian division leader at Schick & Energizer and previously served as the global head of Playtex Infant and Feminine Care. He is a mentor at the University of Toronto’s Hatchery, where he brings deep expertise in consumer goods, sales, and executive strategy. His experience spans global brand leadership and market expansion, making him a valuable advisor on commercialization and business growth.
Aman Thind is the Chief Technology Officer and co-founder of Conavi Medical, a company specializing in catheter-based imaging for cardiology. He provides technical leadership and commercialization insight to Hatchery startups, particularly in the medtech sector. His expertise lies in bridging advanced engineering with clinical application and regulatory pathways.
Charles Boulakia is a partner at Ridout & Maybee LLP, where he specializes in biotechnology, pharmaceuticals, and intellectual property licensing. He advises Hatchery teams on patent strategy, freedom to operate, and regulatory law. His guidance helps early-stage ventures navigate complex IP landscapes and protect their innovations.
Raymond Chik is a semiconductor entrepreneur and co-founder of Untether AI and Kapik Integration. As a mentor at the Hatchery, he supports deeptech founders with expertise in chip design, scaling, and angel investment strategy. His background in hardware innovation and venture building provides startups with critical insights into scaling advanced technologies.
Feedback on Presentation and Revenue Model
Charles Boulakia praised the overall narrative of the presentation but recommended removing certain slides to streamline the pitch. He also advised adjusting the cost model to clearly distinguish between variable and fixed costs. Aman Thind highlighted the need for greater clarity regarding the revenue potential from royalties and suggested that the cost model should reflect how expenses are likely to scale with business growth.
Cost Assessment and Investment Strategy
James Macintosh questioned the relatively low cost estimates presented in the budget, noting the absence of allocations for personnel, marketing, and sales. He advised the team to overestimate costs and prepare a detailed breakdown of potential expenses, including employee salaries. Macintosh also emphasized the importance of presenting best-case scenarios when seeking investment, as investors typically expect significant returns and want to see ambitious growth projections.
Market Potential and Business Model Insights
James Macintosh further discussed the transformative potential of phage therapy in treating bacterial infections, stressing the need for a cohesive narrative that clearly communicates the company’s value proposition. He underscored the importance of quantifying the market size and identifying the number of patients who could benefit from the therapy. Additionally, he expressed confusion regarding the business model, particularly around the number of phages required for effective treatment, and encouraged the team to clarify this aspect for investors.
Communication and Framing of Phage Therapy
Mentors stressed the importance of simplifying the communication of phage therapy for non-technical audiences. Aman Thind and James MacIntosh suggested creating a flowchart to illustrate current processes and future goals. Charles Boulakia provided a concise framing of the problem and solution, emphasizing genetically modified phages as a modern alternative to antibiotics.
Intellectual Property and Business Model
Mentors discussed the distinction between two possible business models. If Mystiphage operates as a tool company, its intellectual property would lie in the optimization process itself, which could be licensed to pharmaceutical companies. These companies would then use the tool to generate multiple drug candidates.
Alternatively, if Mystiphage transitions into a product company, intellectual property would be established the moment a customized protein sequence is created. For example, if the model identifies and optimizes a highly effective protein for MRSA, that protein could be patented as a new drug. In this case, the company would sell the protein to pharmaceutical partners, who would take it through clinical trials and commercialization.
Mentors emphasized that these represent two distinct strategies for monetization and IP ownership, and the team must decide which path to prioritize.
Reflection or Next Steps
Based on feedback from our mentors, we decided to position Mystiphage as a tool company rather than a product company. This approach focuses on providing our AI-driven protein optimization platform to research labs for compassionate use cases, rather than immediately creating patentable phage products. This strategy aligns with the current regulatory landscape, allowing us to operate within established pathways while building credibility and data before pursuing broader clinical applications.
Meeting 4:
Jim Rahaman is a Mentor-in-Residence at the Hatchery, where he guides student startups through financial modeling and strategic planning. With a background in finance and analytics, he previously managed over $500 million in farecard revenues at a major transit agency. His global experience spans consulting, restructuring, and investment cases across 12 countries. Beyond his professional work, he serves on the boards of science education and music organizations, blending his passion for STEM and the arts.
Pratyush Raman is the Go-To-Market Lead at the UofT Hatchery, where he helps startups transition from concept to commercialization. He is the program architect for NEST and LaunchLab, initiatives that have supported over 100 ventures raising more than $100 million. He holds an MBA in Design Thinking from the Rotman School of Management and a degree in Mechanical Engineering from Georgia Tech. His past roles include consultant at EY, engineer at Siemens, and startup advisor, reflecting his passion for technology and impact-driven ventures.
Hope Chik is an Entrepreneurship Mentor at the Hatchery, supporting biotech and health-focused ventures. She is an academic and researcher in medical biophysics, with expertise in stem cells and leukemia. Her cross-disciplinary insight brings scientific rigor to startup ideation and validation, with a focus on translating complex research into viable health innovations.
Regulatory Strategy and Use Cases
The mentors discussed the regulatory landscape for phage therapy. Current applications under compassionate use frameworks target patient-specific phage cocktails, which do not require a formal IND. These are most relevant in markets such as the United States, Poland, and Georgia, where regulatory pathways are more flexible. By contrast, a breakthrough scenario, where phage therapy is positioned as a generalized antibiotic alternative, would require formal regulatory approval, a process that could take up to eight years.
We clarified its positioning as a discovery company rather than a drug maker. Mystiphage provides AI-designed receptor-binding proteins (RBPs), while phage laboratories produce the physical therapies and pursue clinical pathways.
Product and Intellectual Property Insights
Mentors emphasized that while naturally occurring phages are difficult to patent, synthetic AI-designed RBPs may be patentable or licensable. Process IP was identified as a more viable strategy. Our model involves selling in silico sequences to phage labs, which then handle synthesis and application. Validation occurs internally before delivery.
Data privacy and biosecurity considerations were also raised. Receiving patient data would require compliance budgeting of approximately 8,000, with additional costs for data acquisition (5,000), labeling, retraining, and storage.
Go-To-Market and Competitive Advantage
Mentors noted that by providing sequences rather than full therapies, Mystiphage avoids the regulatory burden of drug development. Regulatory costs are indirect and largely borne by phage partners. The company’s uniqueness lies in its AI-generated “hitch proteins,” which improve patient matching by predicting and designing RBPs with greater precision.
However, a competitive scan revealed more than ten companies in the space, underscoring the need for clearer differentiation and refinement of the investor pitch.
Technology and Model Validation
Mentors advised that the “hitch protein” analogy to explain concepts resonates well with audiences and should be used prominently, while the technical term “RBP” can be reserved for appendix slides, which we then incorporated.
Specific Feedback and Communication Strategy
- Jim Rahaman urged the team to establish a clearer conceptual chain, moving from infection to hitch protein, and to test the clarity of this narrative with non-specialists. He also pressed for greater clarity on IP strategy, regulatory costs, partnership dynamics, funding structure, and investor targeting (impact-driven vs ROI-driven).
- Pratyush Raman emphasized the importance of leading with regulatory strategy in pitches, as audiences will expect it. He also encouraged the team to highlight the novelty of their approach within the AI/phage space.
- Hope Chik confirmed that matching accuracy is the perceived value of the technology and asked whether the model directly recommends candidates. She also stressed the importance of distinguishing between in silico design and wet lab validation.
Mentors suggested several slide improvements, including:
- A triangle diagram illustrating the flow from hospital → phage lab → hitch provider.
- Big-picture framing that situates antibiotic resistance among the top three global health risks.
- A visual representation of an RBP (pending validation specifics) to strengthen audience understanding.
Reflection or Next Steps
Based on mentor feedback we clarified the conceptual chain from infection to hitch protein and emphasized our model’s matching accuracy. We also created a one-minute “wow” video summarizing our project process for investors. Moving forward, we will continue refining how we present in silico versus wet lab validation and strengthen regulatory and IP framing in the pitch.
Compliance and Operational Workflow
Wet lab validation would be conducted internally, with Maxwell and Davidson labs handling most documentation, ensuring that no additional regulatory burden falls on the company. The AI-generated sequences are sold to phage labs, which then handle synthesis and regulatory compliance. As such, no IND filing is required by Mystiphage.
The process flow was summarized as: Patient → Hospital → Phage Lab → Mystiphage RBP sequence → Therapeutic solution.
This structure ensures that regulatory responsibility is carried by partner labs, allowing Mystiphage to operate within a low-burden framework.
Meeting 5:
Dr. Charlene Rodrigues is a clinician and researcher affiliated with the University of Toronto, with deep expertise in antimicrobial resistance (AMR), pediatric infectious disease, and global health. Her clinical experience spans both high-resource and low-resource settings, offering valuable insight into the practical realities of treatment deployment, regulatory constraints, and translational innovation.
Phage Discovery and Deployment
Dr. Rodrigues described the traditional phage matching process as akin to “finding a needle in a haystack.” It requires identifying the pathogen, locating or designing a matching phage, producing it, and delivering it to the patient. She expressed optimism about our AI-driven approaches, which she believes could significantly accelerate matching and improve the design of receptor-binding proteins (RBPs).
Adoption Criteria from a Clinician’s Perspective
To consider phage therapy earlier in the treatment pathway, clinicians would require:
- Clinical trial evidence demonstrating safety, efficacy, and appropriate timing
- Clear protocols outlining administration methods and expected outcomes
- Safety data comparable to that required for vaccines and monoclonal antibodies
- Guidelines distinguishing routine use from case-by-case interventions
Dr. Rodrigues emphasized we would need to provide this information should we decide to launch an official startup.
Decision Authority and Cost Considerations
Dr. Rodrigues explained that if phages are stocked and standardized within a hospital, clinicians can administer them freely. However, if phages are externally sourced, expensive, or limited in supply, their use may require approval from a clinical panel. Cost remains a major concern, and many clinicians are currently unaware of the real-world pricing associated with phage therapy.
Business and Translational Considerations
Dr. Rodrigues emphasized that any innovation must be feasible and usable in real clinical settings. She encouraged us to think early about manufacturing pathways, regulatory alignment, and integration into clinician workflows. These considerations are essential for ensuring that promising technologies can transition from bench to bedside.
Meeting 6:
Dr. Alexander Hynes is a leading researcher in phage biology and a founding member of Phage Canada. His work spans microbial genetics, synthetic biology, and translational phage therapy, with a particular focus on the intersection of scientific innovation and commercialization. As both a scientist and a realist, Dr. Hynes brings a critical perspective to the financial and regulatory challenges facing phage-based therapeutics and offered us candid insights into what it will take to move the field forward.
The Financial Case for Phage Therapy
Dr. Hynes began by drawing a parallel between antibiotics and fire extinguishers, essential tools that must be available but are ideally used sparingly. This paradox, he explained, undermines the financial viability of antibiotic development. Even companies that successfully brought new antibiotics to market since the 1980s have gone bankrupt due to insufficient sales to recoup R&D costs.
Phage therapy, he argued, faces an even steeper financial hill. Although phage development and production are generally cheaper, the challenges of quality control, regulatory uncertainty, narrow host ranges, and short shelf life make them commercially fragile. Even if phages prove effective, their rare usage limits revenue potential.
Funding Models and Economic Innovation
To address the economic shortcomings of antibiotic development, Dr. Hynes pointed to the UK’s adoption of a “subscription model,” where governments pay for access to antibiotics regardless of usage, similar to an insurance policy. This model helps stabilize revenue streams and incentivize innovation.
He suggested that phage therapy could benefit from similar frameworks, though acknowledged that the financial case remains tenuous. The core issue, he emphasized, is not efficacy but economics: who will pay for phage therapy, and under what conditions?
Regulatory Progress and Investor Hesitation
Dr. Hynes noted that several countries (including Belgium, Portugal, the UK, and Australia) have eased regulations around phage administration. The U.S. FDA has shown support through the IND and emergency IND frameworks, and Canada is exploring models akin to those used for fecal transplants.
Despite these regulatory advances, investor capital has not followed. Even in Belgium, where hundreds of patients have received phage therapy under the new system, the lack of a clear return on investment has kept private funding at bay.
Where the Money Is (and Isn’t)
Dr. Hynes acknowledged that some companies are finding viable business models by diversifying beyond human therapeutics. He cited Cytophage, a Canadian company that went public and earns most of its revenue from agricultural applications, despite gaining media attention for human cases.
Agriculture offers several advantages:
- Large, genetically uniform populations
- Predictable outbreaks
- Fewer regulatory restrictions
- Clear economic incentives
Phages can be used prophylactically in agriculture (e.g., spraying fields before an outbreak), which is not permitted for antibiotics. The low production cost and scalability in high-volume environments like poultry farms make agriculture a more commercially viable sector.
Clinical Viability and Therapeutic Scope
When asked about standardized phage treatments for conditions like urinary tract infections (UTIs), Dr. Hynes explained that while E. coli is a common cause and well-studied, its genetic diversity makes one-size-fits-all phage cocktails unlikely. Clinical trials tend to focus on conditions with expensive standard treatments—such as inflammatory bowel disease (IBD)—to justify the high price point. UTIs, by contrast, are relatively inexpensive to treat with antibiotics, even when recurrent.
Synthetic Biology and Host Range Expansion
Dr. Hynes expressed cautious optimism about synthetic biology’s potential to overcome phage specificity challenges. Modular phage design, such as swapping tail fibers or receptor-binding proteins (RBPs), is an active area of research. While de novo RBP design is rare, it is not unimaginable, especially with tools like AlphaFold.
However, he noted that current datasets for training AI on protein-glycan interactions are limited and lack standardization, which constrains progress. Most efforts focus on swapping existing modules rather than designing RBPs from scratch.
Commercial Bottlenecks and the Path Forward
Dr. Hynes emphasized that the key bottleneck in phage therapy is not technical but commercial. Without a viable financial model, scientific progress will remain dependent on grant funding and move slowly. He drew a comparison to oncology, where CAR-T cell therapies transitioned from theoretical to real-world treatments because the economics supported it.
September
Learning about the Microbiome and Targeting the Gut Resistome with Phage Therapy
During the opening days of September, the issue the dry lab team was tackling was how to translate computationally designed phages into a realistic therapeutic strategy for the gut resistome, while avoiding overengineering solutions that were not biologically necessary. Furthermore, we also had to tackle whether there will be any collateral/background selection when dosing phage orally vs. directly at the target site with an electroceutical pill.
Dr. Bryan Coburn
To answer these questions, we consulted with Dr. Bryan Coburn (MD, PhD, FRCPC), whose work focuses on how antibiotics, pathogens, and the human microbiome interact. He leads clinical trials in infectious diseases, antimicrobial stewardship, and microbiome-targeting interventions, making him uniquely positioned to evaluate how phage therapy could be integrated into real-world patient care (5).
Effects of Antibiotic Use on the Gut Microbiome
Dr. Coburn explained that antibiotic use in both humans and animals exerts profound effects on the gut microbiome. Antibiotics select for antimicrobial-resistant (AMR) organisms, kill off beneficial anaerobes, and create ecological niches that favour pathogens. He highlighted that there is often a positive correlation between pathogen density and the presence of antimicrobial resistance genes (ARGs). For our workflow, this underscores the importance of designing phage interventions that account for the disrupted microbial ecology left behind by antibiotics. Rather than assuming a neutral baseline, we need to model and test phage efficacy in the context of a microbiome already skewed toward resistance and pathogen dominance.
Viability of Targeting the Gut Resistome
Dr. Coburn was enthusiastic about the viability of targeting the gut resistome as a strategy to slow the development of antimicrobial resistance. He described it as an “excellent” approach, particularly in animals, where the gut resistome is both a diagnostic and therapeutic target. He emphasized the appeal of speed in this context, noting that rapid interventions that can eliminate resistant organisms before they spread are highly valuable. He also noted that many humans carrying AMR organisms naturally decolonize within about six months, suggesting that interventions should focus on high-risk windows rather than universal decolonization. He pointed to the SER-155 trial, which used a microbial consortium to decolonize transplant patients, as a proof-of-concept for this type of strategy. For our workflow, this means we should frame phage therapy not as a universal decolonization tool, but as a targeted, time-sensitive intervention for high-risk populations or livestock systems where speed and efficiency matter most.
Challenges in Delivering Phage Therapies to the Gut
When asked about challenges in delivering and maintaining effective phage therapies in the gastrointestinal tract, Dr. Coburn did not identify specific barriers. However, his broader comments suggest that overcomplicating delivery mechanisms may not be necessary. For our workflow, this implies that we should prioritize robust, scalable delivery methods (such as oral dosing or encapsulation) rather than investing heavily in complex, conditional release systems unless there is clear evidence of added benefit.
Collateral Selection and Oral Dosing
Dr. Coburn addressed concerns about collateral or background selection when dosing phages orally versus using targeted release devices. He argued that we should not begin by trying to fine-tune phage release at specific gut loci. If the goal is to kill a pathogen like E. coli, some off-target effects are acceptable in the short term. This advice suggests that our workflow should not over-engineer precision delivery systems at the expense of speed and scalability. Instead, we should focus on demonstrating efficacy in clearing pathogens, even if collateral effects occur, as long as they are tolerable and do not destabilize the microbiome long-term.
Conditional Release Mechanisms
On the question of whether bacterial infections move around in the gut and whether this justifies conditional release mechanisms, Dr. Coburn was clear: such strategies are dispensable. He explained that organisms generally occupy the same anatomical niche, so conditional release does not provide meaningful protection for other organisms. He noted that while this is a big topic in cancer therapy, it is not directly relevant for our case. For our workflow, this meant we deprioritized conditional release mechanisms and instead focused on simpler, more direct dosing strategies that can be validated quickly.
Biofilm and Quorum Disruption
When asked about biofilm or quorum disruption as adjunctive strategies, Dr. Coburn admitted that he did not have a definitive answer. He noted that small intestinal bacterial overgrowth (SIBO) is a specific syndrome, but not the type of problem we are describing. His uncertainty here suggests that while biofilm disruption may be an interesting adjunct, it should not be a central focus of our workflow until more evidence emerges.
Generalizable Pill Devices for Gut Decolonization
Dr. Coburn mentioned that a generalizable pill device that could prune the gut resistome of any animal could have relevance if able to be deployed at low cost and at high throughput. He also pointed out that there is already a whole field working on this, with fecal microbiota transplantation (FMT) now being encapsulated for agricultural use. He suggested that the colon is the most relevant target, but emphasized that the first question should always be: can you decolonize? If the answer is no, then no delivery mechanism, no matter how sophisticated, will solve the problem.
Reflection or Next Steps:
We validated that a generalizable electroceutical pill as a targeted intervention for gut resistome modulation is useful, and deprioritized applications for it that were not as useful. Dr. Coburn’s guidance reinforced that the first priority is demonstrating efficacy in reducing pathogen load, rather than overengineering delivery mechanisms, which aligned with our focus on the pill’s direct, measurable impact.
Bridging Computational Generative Models with Biological Relevance in Predictive RBP Design
Andrei Bogza (PhD Student)
At this stage of our pipeline, our hardware team was tackling the misalignment between our computational design focus and the added complexity of hardware-based delivery systems like capsules. To answer our concerns, we consulted with Mr. Andrei Bogza, an emerging voice in microbiology and immunology whose research sits at the intersection of host–pathogen interactions, bacterial stress responses, and the microbiome. Trained in infectious disease biology, his work has explored how pathogens such as Salmonella adapt to hostile immune environments, including the role of phage shock protein systems in resisting antimicrobial peptides (6).
Feasibility of the Hardware Pill Idea
Bogza expressed skepticism about the feasibility and necessity of a hardware pill for phage delivery. He noted that while phages targeting conditions like small intestinal bacterial overgrowth (SIBO) may benefit from targeted delivery, the first action of phage therapy will almost always occur in the small intestine, regardless of the delivery method. Capsules could be useful in cases where therapy must bypass the small intestine and reach specific sites like the colon, but these are rare and highly complex applications, such as in ulcerative colitis or Crohn’s disease. He emphasized that avoiding stomach acid is a valid concern, but this can be solved with a much simpler approach: administering higher phage doses, since phages are not toxic even at concentrations of 10^8^-10^10^ PFUs/mL.
Identifying the Best Target Sites
When asked about the most appropriate target sites for phage delivery, Bogza suggested that the colon or distal ileum would be the most logical choice. From a storytelling and translational perspective, the strongest case would be if we could show that depleting bacteria at the site of interest is beneficial, while off-target depletion is detrimental. However, he cautioned that this narrative may be difficult to support given the limited research on microbiome dynamics outside of the colon. For our workflow, this means that while site-specific targeting is conceptually appealing, we should be cautious about overpromising precision in areas where the science is still underdeveloped.
Targeting the Gut Resistome
On the question of whether targeting the gut resistome makes biological sense, Bogza admitted that this was somewhat outside his domain. He recalled that groups at the University of Guelph have explored phage therapy for veterinary purposes, including related ideas, but he was not aware of specific research linking AMR spread directly to wildlife gut resistomes. Instead, he pointed to antibiotic runoff into water systems as a more established driver of subtherapeutic antibiotic exposure and resistance spread. He also reiterated his doubts about the usefulness of capsules in this context, noting that gut sequencing reveals highly variable dynamics across different regions, with strain-level differences and lower biomass in the small intestine compared to the colon. His advice was that rather than investing in complex capsule systems, we could use simpler, pH-sensitive microencapsulation devices to protect phages through the stomach and release them in the colon. This approach is more aligned with the realities of gut ecology and avoids the pitfalls of overengineering.
Turning Bacteria On or Off in the Gut
He described the microbiome as a jungle where all bacteria are mixed, making it rare that one would need to deliver phage therapy to a very specific site. In his view, general delivery is more than sufficient for most applications. This reinforces his broader message: the complexity of the microbiome does not necessarily demand equally complex delivery mechanisms. Instead, the focus should be on ensuring that phages reach the gut in viable form and at sufficient doses, rather than on engineering highly specific release devices.
MAIN TAKEAWAYS:
Based on Bogza’s advice, our workflow should avoid overengineering hardware pills, recognizing that most phage activity occurs in the colon and distal ileum, and found that high-dose oral delivery is a simpler, effective solution to bypass stomach acid degradation. We plan to explore pH-sensitive microencapsulation approaches as a tractable alternative to more complex capsule systems. This direction aligns with gut ecological constraints and avoids unnecessary formulation complexity. Additionally, rather than attempting selective “on/off” bacterial control—which remains challenging given the microbiome’s inherent variability—we intend to focus on delivery designs that maintain broad therapeutic efficacy. Our long-term goal is to enhance viable phage dosing and survival within the gut environment before moving toward more sophisticated, conditionally responsive release mechanisms.
Final Reflections and Takeaways:
Targeted Protein Design vs. Genome-Wide Mutation
At our final stages of our project, we needed to tackle how we were going to differentiate and justify our protein-level design approach compared to genome-wide modification strategies.
Dr. Bin Shao:
To tackle this issue, we consulted with Dr. Shao, President of the Zhongguancun Institute of Artificial Intelligence and a full professor at Zhongguancun Academy, whose expertise lies in artificial intelligence, computational biology, and scientific computing. He founded the Graph Engine Project and established both the Computational Biology and Scientific Computing Research Groups. He has more than fifty publications in Nature journals and top conferences, along with awards such as the ICDE 10-Year Influential Paper Award. His team’s breakthroughs in protein dynamics, recognized among the Top 10 Advances in Chinese Bioinformatics in 2024, are directly relevant to fields like phage therapy, immunology, and infectious disease, where understanding protein interactions is critical.
Overall Assessment of the AI Approach
Dr. Shao was supportive of the core direction of our AI‑generative, predictive RBP project, describing the approach as clever and highlighting the active learning component as particularly exciting. He noted that leveraging pre‑trained protein language models such as ESM is a strong foundation, but advised that fine‑tuning these models would significantly improve their performance for our specific application. His advice here was clear: while using pre‑trained models is a good starting point, our workflow should prioritize customizing them to our dataset and objectives to maximize predictive accuracy and biological relevance.
Genome vs. Protein Design Approaches
A central theme of the discussion was the distinction between genome‑wide mutation approaches (such as those explored in recent Arc Institute work) and our protein‑focused RBP design strategy. Dr. Shao pointed out that genome‑based methods often introduce 50–300 mutations, many of which occur in non‑coding regions, raising the question of whether such phages are truly “generated” or simply “modified.” While some of these mutations may affect RBPs, the changes are diffuse and less targeted.
By contrast, he emphasized that our workflow’s protein‑centric approach offers a clear advantage: it enables precise, interpretable, and flexible design of RBPs, directly tied to binding specificity and therapeutic function. Dr. Shao emphasized that this targeted design is more practical in the near term, whereas genome‑wide approaches may only become viable much further into the future. His advice reinforced that we should lean into the narrative of precision and flexibility.
Demonstrating Model Soundness Without Experimental Data
One of the biggest challenges we face is how to validate and showcase the credibility of our AI pipeline before experimental results are available. Dr. Shao offered several concrete suggestions for strengthening our workflow presentation:
- Highlight computational outputs: Show predicted differences in binding affinity and demonstrate how proteins were optimized.
- Demonstrate breadth: Present results across multiple bacterial targets to prove generalizability.
- Pair AI‑RBP predictions with receptor binding data: Showcase side‑by‑side evidence of how designed RBPs interact with bacterial receptors.
- Be transparent about active learning: Acknowledge that active learning can be tricky, but emphasize how it improves model robustness over time.
- Leverage the phage bank: Present the size, diversity, and structure of our phage bank as evidence of scalability and systematic design.
This advice points directly to a workflow issue we were tackling: how to build credibility and trust in our AI outputs before wet‑lab validation.
Information Needed to Judge Project Strength
From his computational perspective, Dr. Shao outlined the types of information that would help him (and by extension, other computational experts) evaluate the strength of our project. These included:
- A complete pipeline: Demonstrating end‑to‑end integration, from data input to RBP design and validation.
- Multiple ESM models: Exploring the use of two different fine‑tuned models to compare performance and robustness.
- Binding affinity predictions in nanomolar (nM) ranges: Providing quantitative, interpretable metrics that align with experimental standards.
- Docking experiment results: Visual and structural evidence of RBP–receptor interactions.
Phage bank details: Clear data on the size, diversity, and design principles of our phage library.
Targeted Protein Design vs. Genome-Wide Mutation
We also needed feedback on how to build a rigorous evaluation framework that goes beyond simple binding predictions to ensure safety and structural validity.
Dr. Brett Trost:
To answer our questions we consulted Dr. Brett Trost, a computational biologist and Assistant Professor in the Department of Molecular Genetics at UofT, as well as a Scientist in the Molecular Medicine program at SickKids. His research focuses on leveraging multiomic data, integrating genomics, transcriptomics, proteomics, and epigenetics, with advanced machine learning to uncover the molecular underpinnings of complex diseases such as autism. Widely recognized for his impact, Dr. Trost has received prestigious awards including the Governor General’s Academic Gold Medal and the CIHR Banting Postdoctoral Fellowship, and was invited to the Lindau Nobel Laureate Meeting. With all these accolades, we knew he was the perfect stakeholder for our workflow.
Filtering for Desirable Phage Properties
One of the first issues raised in the discussion was how to ensure that only safe and therapeutically viable phages are advanced through the pipeline. Dr. Trost endorsed the predictive filtering step in our phage bank, which screens out candidates with undesirable properties such as lysogeny or toxin genes, calling it a “smart way of doing that.” His advice reinforced the importance of embedding safety filters early in the workflow, so that downstream design and optimization efforts are focused only on phages with clinical potential. For our project, this means continuing to refine and expand the predictive screening layer, ensuring it is robust enough to catch not only obvious red flags but also subtler genomic features that could compromise therapeutic safety.
Assessing the Success of Generated RBPs
Dr. Trost emphasized that evaluating RBP designs cannot rely on binding predictions alone. He highlighted the need for a multi‑layered assessment strategy that integrates structural and functional validation. This includes using metrics such as ipTM and pLDDT scores to assess structural confidence, docking simulations to evaluate receptor interactions, and qualitative “vibe checks” to ensure that the protein design makes biological sense. His advice pointed to a workflow gap: while our AI pipeline can generate candidate RBPs, we must strengthen the validation stage with diverse computational checks that collectively build confidence in the plausibility and functionality of each design. Integrating these layers into a standardized evaluation framework would make our outputs more credible to both computational and experimental stakeholders.
Expanding Functionality: Biofilm Degradation
When asked about whether RBPs could be engineered to degrade biofilms, Dr. Trost supported the idea of enhancing functionality by engineering tail spike proteins (TSPs) with enzymatic activity. This advice is particularly relevant because biofilms represent one of the most persistent barriers in treating chronic infections, shielding bacteria from both antibiotics and phages. For our workflow, this means considering functional augmentation as part of the design process, not just optimizing RBPs for binding, but also exploring modular engineering strategies that add enzymatic or biofilm‑disrupting capabilities. Incorporating this into our AI pipeline could differentiate our project by addressing a major clinical challenge that standard phage therapy often struggles with.
Dr. Garton
Dr. Garton’s expertise lies in biomedical engineering and synthetic biology. As a Canada Research Chair in Synthetic Biology, his lab focuses on integrating generative AI, protein design, and stem cell engineering to create next‑generation gene and cell therapies.
His insights helped us clarify how our workflow might be interpreted by regulators, emphasizing the importance of explicitly addressing spread risks, prioritizing wet-lab validation, and maintaining a lytic-only design.
Regulatory Concerns and the Risk of Spread
Dr. Garton raised an important issue around the regulatory framework for engineered phages, particularly the possibility of unintended spread. He noted that phage genomes can sometimes integrate into bacterial genomes, and if those bacteria spread to another person, the engineered phage could effectively be transferred as well. Even if our workflow restricts itself to strictly lytic phages (which do not integrate into host genomes) he cautioned that the human microbiome is so diverse that it would be impossible to pre‑screen against every possible bacterial host. In practice, this means that patients treated with engineered phages might need to be highly isolated to prevent unintended ecological effects. He made it clear that our workflow must anticipate regulatory scrutiny by explicitly addressing containment and spread risks, and we should consult experts (ex: Prof. Lan) to strengthen our safety framework.
The Importance of Testing and Validation
Dr. Garton emphasized that testing is the critical step that transforms a project from theoretical promise into credible science. Speaking from experience, he explained that a single piece of data showing that a designed phage can bind to a new bacterial target is what convinces both regulators and the scientific community that the approach is real. For our workflow, this means that computational predictions alone are not enough, we must prioritize generating wet‑lab validation data that demonstrates binding, specificity, and safety. His advice pushes us to integrate a test‑driven development mindset into our pipeline: every computational design should be paired with a clear experimental plan for validation.
Safety Considerations for Human Practices (HP)
For the Human Practices safety section, Dr. Garton outlined three key safeguards that should be explicitly built into our workflow and communications:
- Lytic‑only phages: ensuring that all phages in the bank are strictly lytic, eliminating the risk of genome integration.
- Broad bacterial testing: validating engineered phages against an array of bacterial strains in the wet lab before any patient use, to minimize off‑target risks.
- Agricultural applications: for use cases like phage pills in livestock or crops, emphasizing the large array of testing and the lytic nature of the phages to reassure regulators and the public.
Possible leap into Agriculture
Clark Sinclair
Furthermore, we were curious about the role of antibiotics in the role of farming. To answer our question, we consulted with Clark Sinclair, a veterinarian who specializes in cattle/large animal medicine. Raised in Acton, Ontario, Clark developed an interest in animals and agriculture from a very early age. It began on his grandparents’ farm and grew steadily from there, with cattle always being his primary focus. His passion for agriculture and livestock led him to the Ontario Agricultural College, where he earned a Bachelor of Science in Agriculture in 2010. After completing his undergraduate degree, he sought a career that would allow him to work directly in the dairy industry and make a meaningful impact at the farm level. His main interests include dairy production medicine, health management, general medicine, and surgery. He also occasionally works with beef cattle.
Challenges in Treating Bacterial Infections
Clark believes antibiotic resistance is a growing concern for both veterinarians and farmers. While he doesn’t encounter resistance daily, he does see it in specific cases. Notably, about 90% of the time, animals are initially treated by farmers—typically with antibiotics—and veterinarians are called in only when those treatments fail. When he steps in, he often performs sensitivity tests on bacterial cultures. Although he hasn’t yet encountered a bacterium that is completely resistant to all antibiotics, he has seen strains resistant to up to 80% of available options.
Euthanasia and Untreatable Infections
Some animals die from septicemia even while receiving antibiotics. However, it’s difficult to determine whether these outcomes are due to antibiotic resistance or simply because the disease had progressed too far by the time treatment began.
Antibiotic Use in Farming
Antibiotics are primarily used to treat diseases. Other applications include ionophores, which modify the cow’s rumen for specific purposes, and prophylactic treatments to prevent infection when cows enter high-risk environments. Although vaccines are used, they do not fully replace the utility of antibiotics.
Reducing Antibiotic Use: Is It Practical?
Reducing antibiotic use is possible, but there will always be cases where treatment is necessary. Even with improved farming techniques and preventative measures, some animals will inevitably fall ill. Clark has observed a reduced need for antibiotics as farming practices evolve and believes the industry is moving in the right direction. However, complete elimination of antibiotic use is unrealistic.
Perceptions of Our Product
Clark sees potential in the product but emphasizes that it must be significantly streamlined. Since farmers handle the majority of first-line treatments, the product must be easy to use, cost-effective compared to antibiotics, and practical—ideally requiring minimal intervention. For example, waiting for lab results could be fatal for a cow, so immediate treatment options are essential. While phages offer specificity, this could be a drawback; broad-spectrum phages would be more useful in urgent scenarios.
Post-COVID Perceptions of Viral Therapies
Perceptions of viruses and biopharmaceuticals have shifted since COVID. Some farmers and vets may be receptive to phage therapy, while others remain cautious. They would need assurance that the treatment won’t cause adverse effects, such as illness or death. Although antibiotics have known side effects, there is concern about what side effects viral therapies might introduce.
AI Adoption in Farming
Some early adopters would be open to using AI-based tools, but there is skepticism around big data in certain farming communities. If the product proves effective among early users, it could help convince others. With increasing technological innovation in agriculture, many farmers are open to new tools—but they require strong evidence of efficacy.
Marketing to Farmers
Cost is the most critical factor, if the product isn’t economically viable, adoption won’t happen. Effectiveness and practicality are also essential. One promising angle is reducing the number of injections required during antibiotic courses. If the product allows for a single treatment, it could fill a valuable niche. The pill-based design is a major advantage, as farmers generally dislike injecting large animals. Ideally, the treatment would be administered just once via feed.
Integrated Human Practices for our Wiki:
Dr. Jovana Drinjakovic
We also conducted integrated human practices for our own Human Practices wiki! To get feedback about Wiki drafts, we interviewed Dr. Jovana Drinjakovic, a Scientific Communications Officer at University of Toronto. She specifically specializes in translating complex scientific concepts into accessible narratives for diverse audiences.
Project Highlights for Public Communication
Dr. Drinjakovic advised that the project should emphasize the role of viruses, particularly in the context of public perception following the COVID-19 pandemic. She recommended acknowledging the fear many people still associate with viruses, while clarifying that the viruses used in this project are completely safe for human use. The messaging should stress that these viruses only infect bacteria, not humans, and that they are being harnessed as a powerful tool for medicine.
Anticipated Public Misunderstandings
According to Dr. Drinjakovic, the greatest challenge for a lay audience will be understanding technical terms such as “bacterial surface proteins,” “how phages see bacteria,” and “how they enter bacteria.” She cautioned against delving too deeply into molecular details, as this may overwhelm or confuse non-specialists. Instead, she recommended using analogies to simplify concepts. For example, comparing bacteria to houses with postal addresses that phages can recognize and follow. She suggested avoiding unnecessary jargon and focusing on clear, relatable explanations.
Recommended Formats for Public Engagement
Dr. Drinjakovic emphasized the importance of visual communication. She recommended using diagrams, infographics, and illustrations to convey complex ideas in a digestible way. These formats, she noted, are particularly effective for public audiences who may not have a scientific background but can quickly grasp visual representations of processes and concepts.
Feedback on Project Graphics
Regulatory Framework Graphic
Dr. Drinjakovic recommended changing the phrase “different places” to “around the world” for greater clarity and inclusivity in the regulatory framework. She also noted that the use of all capital letters should be avoided, as it is not accessible and does not align with best practices. Overall, however, she found the design strong and well executed.
Newsletter
She praised the newsletter as “very beautiful” and highlighted the inclusion of the history of phage therapy as a valuable addition. She recommended aligning the paragraphs to the left for improved readability and professional presentation.
Social Media Post on Bioethical Implications
Dr. Drinjakovic cautioned against beginning a public-facing post with technical terminology such as “phage therapy.” Instead, she recommended framing the content with accessible, thought-provoking language, such as: “Is using viruses as a medicine ethical?” This phrasing, she explained, would capture the attention of a broader audience. From there, the post should link viruses to antibiotic resistance, for example: “Viruses can help address one of the biggest health problems of our time.” Only after this framing should antibiotic resistance be introduced. She noted that the visuals accompanying the post were strong and engaging.
Problem Statement
Dr. Drinjakovic recommended strengthening the problem statement with compelling statistics. She suggested referencing historical data on how deadly bacterial infections once were, including the percentage of patients who would die without recovery, and emphasizing that bacteria will once again become deadly if new solutions are not found. She advised framing phages as an untapped resource and highlighting the project’s role in precision medicine, particularly given the limited diversity of complex molecules currently available.
She also stressed the importance of introducing viruses before using the term “phages,” to ensure clarity for audiences unfamiliar with the terminology. Receptor-binding proteins may need to be explained in simple terms, and technical references such as the WHO priority list should be spelled out explicitly, for example, clarifying that “people are dying because antibiotics are no longer working.” Finally, she encouraged the inclusion of alarming statistics on antibiotic resistance to add urgency and strengthen the case for the project.
October
Dr. Evelien Adriaenssens
At the start of October, we sought out Dr. Adriaenssens’ expert perspective on the scientific and regulatory challenges of phage therapy, particularly regarding receptor-binding protein (RBP) design, infection dynamics, and the feasibility of AI-driven phage discovery. The goal was to validate Mystiphage’s dry lab pipeline and explore its translational potential.
Key Insights from the Interview
RBP Design & Model Assumptions: Dr. Adriaenssens emphasized the importance of clearly defining design assumptions, such as focusing solely on RBPs and relying on accurate bacterial receptor predictions. She noted that glycan-based targeting is novel and promising, but receptor complexity (e.g., sugar modifications) adds layers of challenge.
Phage Diversity & Infection Dynamics: She highlighted that phage infection is not solely determined by adsorption and injection. Resistance can arise from multiple bacterial defense systems, many of which remain poorly understood. Even genetically similar phages can behave differently, and predictions may fail without deeper biological context.
Model Validation & Wet Lab Testing: Dr. Adriaenssens supported the our use of E. coli phages Mu and P2, noting that starting with known glycan-binding phages is practical. She stressed the importance of validating infection cycles experimentally and acknowledged the complexity introduced by larger phage genomes.
Regulatory & Commercial Barriers: She explained that while phage therapy is used in countries like Poland and ex-Soviet states under compassionate use, broader adoption requires standardized clinical trials with fixed cocktails. Regulatory hurdles and lack of pharma investment remain major obstacles.
AI & Future Potential: Dr. Adriaenssens expressed cautious optimism about AI’s role in personalized phage design, provided it’s used responsibly. She sees potential in building scalable, low-cost production pipelines and integrating AI with clinical workflows but emphasized that commercial viability must be solved to unlock broader impact.
Professor Artem Babaian
We also met with Professor Babaian to explore the scientific and computational foundations of Mystiphage’s AI-driven approach to phage therapy. The goal was to evaluate whether generative models could offer meaningful advantages over traditional bioinformatics and phylogenetic sampling in designing receptor-binding domains (RBDs) for phages.
Professor Artem Babaian is a molecular biologist known for his work in viral genomics and computational biology. He brings a critical and evolutionary perspective to synthetic biology, with particular interest in how natural diversity can inform design strategies.
Key Insights from the Interview
- Phage Therapy Challenges: Professor Babaian acknowledged the promise of phage therapy but emphasized current barriers—high screening costs, regulatory complexity, and bacterial resistance.
- AI vs. Bioinformatics: He questioned the need for AI, suggesting that large-scale bioinformatics and sequence similarity methods might suffice. We responded that AI could narrow down viable RBP candidates more efficiently than brute-force sequence comparisons.
- Phylogenetic Sampling: Babaian recommended sampling proteins from adjacent clades in the phage phylogenetic tree rather than designing new sequences from scratch. He argued that many useful proteins likely already exist in nature but remain undiscovered due to limited sequencing coverage.
Last Insights
Finally, to end off October, we conducted an interview with Professor Christopher Yip, the Dean of the Faculty of Applied Science and Engineering, to get his final insights:
Professor Christopher Yip
Professor Christopher Yip is the Dean of the Faculty of Applied Science & Engineering at the University of Toronto, where he leads one of the world’s top-ranked engineering faculties. He holds appointments in the Department of Chemical Engineering & Applied Chemistry and the Institute of Biomedical Engineering, with cross-affiliations to the Donnelly Centre and the Department of Biochemistry in the Temerty Faculty of Medicine.
He emphasized two main points:
Modeling Receptor Variability
Dr. Yip underscored the complexity of glycan receptors, noting that their conformations are highly flexible and vary depending on environmental conditions and bacterial state. To address this variability, he recommended that robust generative models should:
- Utilize conformational ensembles derived from molecular dynamics (MD) or nuclear magnetic resonance (NMR) data, rather than relying on static receptor structures.
- Train on diverse receptor datasets and apply data augmentation techniques to improve generalizability.
- Incorporate feedback from experimental validation platforms such as glycan arrays and cryo-electron microscopy (cryo-EM).
- Combine physics-informed constraints with machine learning adaptability to ensure both accuracy and flexibility in design.
This multi-layered modeling approach would allow Mystiphage to better capture the dynamic nature of bacterial receptors and improve the precision of RBP targeting.
Research Gaps and Strategic Applications
Dr. Yip identified several unresolved challenges in phage system development, including limited understanding of mutation drivers, selection pressures, and variability during scale-up. He suggested that Mystiphage could play a pivotal role in addressing these gaps by mapping mutation pathways and linking computational predictions to experimental evolution. This would enhance the project’s holistic understanding and improve its translational potential.
He also pointed to broader applications of phage technology, such as bioremediation, which demonstrate the scalability and environmental relevance of the platform. These use cases could help position Mystiphage not only as a therapeutic innovator but also as a contributor to sustainable biotechnology.
References
- Strathdee S. 2025. Profile: Steffanie Strathdee. UC San Diego. https://profiles.ucsd.edu/steffanie.strathdee
- German G. 2025. Faculty Profile: Laboratory Medicine and Pathobiology. University of Toronto. https://lmp.utoronto.ca/faculty/greg-german
- Piuri M. 2025. Loop Profile: Mariana Piuri. Frontiers Research Network. https://loop.frontiersin.org/people/541661/overview
- Parent K. 2025. Faculty Profile: Kristin Parent. CryoEM Center for Bioanalysis and Computation (CCBC). https://www.ccbc-efrc.org/copy-of-karen-mulfort
- Coburn B. 2025. Faculty Profile: Bryan Coburn. Department of Immunology, University of Toronto. https://immunology.utoronto.ca/faculty/bryan-coburn
- Massicotte R, et al. 2025. The Salmonella phage shock protein system is required for defense against host antimicrobial peptides. PLoS Pathog. 21(9):e1013132. https://journals.plos.org/plospathogens/article?id=10.1371/journal.ppat.1013132
- Asekomhe IF. 2025. Physician Profile: Dr. Ikhimhiagie Felix Asekomhe. HealthDoc Canada. https://www.healthdoc.ca/listing/dr-ikhimhiagie-felix-asekomhe-076dbd