Overview

As iGEMers, we are committed to contributing to the iGEM community by sharing our project experiences, newly developed biological parts, computational models, and human practices materials. Our wet lab team has created several novel biological parts, while the dry lab team has successfully developed multiple computational models, including a de novo RNA aptamer generation model, an antibody directed evolution model, and a model describing the kinetics of OMVs and "Expiry Date Circuit", along with innovative experimental protocols. Beyond laboratory work, we actively participated in collaborations and documented our innovative concepts in accessible articles to benefit other teams. Furthermore, we established a comprehensive Integrated Human Practices (IHP) methodology to deeply explore the connections between our project and real-world therapeutic applications.

Contributions to Parts

HemorrEaser incorporates a novel bioPROTAC (VHH-VHL) designed to mediate ubiquitination-mediated degradation of non-hydroxylated HIF-1α under hypoxic conditions, thereby inhibiting angiogenesis in hemorrhoidal tissues. The VHH-VHL construct consists of four components: a nuclear localization signal (NLS), an HIF-1α-binding nanobody (VHH212), an E3 ubiquitin ligase recruiter (pVHL), and linker peptides.

Functional validation was performed through integrated dry and wet lab approaches. Computational assessments included molecular docking (HADDOCK/HDOCK) of the HIF-1α-bioPROTAC-ubiquitin complex, Amber molecular dynamics simulations, and Gaussian 16-based transition state analysis. Results indicated a feasible ubiquitination energy barrier of ≈8.20 kcal/mol, supporting reaction viability.

Experimental validations confirmed that the VHH-VHL construct maintains intact binding to HIF-1α, as demonstrated by yeast two-hybrid and bacterial co-purification assays. Furthermore, in vitro ubiquitination assays specifically confirmed HIF-1α polyubiquitination in the presence of bioPROTAC, collectively verifying its dual functionality.

Collectively, these results validate the rational design of VHH-VHL as an effective degrader of HIF-1α under hypoxia, providing a mechanistic basis for its anti-angiogenic function in hemorrhoid therapy.

For details, refer to the Part: BBa_25U91GJF

We deposited in the iGEM Parts Registry an anti-VEGF monovalent nanobody carrying a masking peptide. P17 was fused to the N-terminus to enhance solubility; a masking peptide connected via an MMP-3-cleavable linker was appended to the C-terminus to occlude the antigen-binding site. Within the hemorrhoidal microenvironment, MMP-3 cleaves the linker, releasing the mask and exposing the VEGF-binding interface. This permits local, selective VEGF blockade, disrupts the VEGF-VEGFR signaling axis, and suppresses pathological angiogenesis. The part provides the iGEM community with a low-risk, hemorrhoid-targeted antibody reagent.

For details, refer to the Part: BBa_2546SCTX

For the precise treatment of hemorrhoids, we have selected Escherichia coli outer membrane vesicles (OMV) this year as the delivery platform for the anti-vascular module. OMV has great potential as a drug delivery system, and we have chosen the OmpA-CAR fusion protein as the targeting element based on literature. On one hand, the α-helix of the OmpA signal peptide can efficiently insert into the membrane structure; on the other hand, CAR can specifically recognize the angiogenic endothelial cells abundant in the Hemorrhoids site. Achieving the functional objective of membrane anchoring-specific recognition, it can effectively mediate the targeted transport of OMV to the hemorrhoids site.

For more details, please refer to: BBa_25WEXRPG

We designed a QS-coupled TetR-mediated feedback suppression element using the LuxI, LuxR, and TetR modules, which have been fully characterized in the Registry. The system enables autonomous density-dependent regulation: constitutive luxI expression produces AHL, which accumulates with cell density and activates LuxR to drive tetR expression from the P_lux. TetR subsequently represses downstream gene expression via P_tet, establishing a closed-loop control mechanism that maintains expression stability without external intervention. The specific regulatory kinetics of this circuit are shown in the Figure 5., reflecting the downstream gene expression level per unit of engineered bacteria during the stationary phase. This feedback circuit provides a tunable, self-regulating module for synthetic consortia, allowing precise control of therapeutic gene expression in response to population changes. It serves as a foundational tool for engineering robust microbial systems in biomedical and bioprocessing applications.

For details, refer to the Part: BBa_25U91GJF

We proposed the "medicine-food collaboration" paradigm and experimentally validated it using a riboswitch. We verified that under TPP induction, the thiM riboswitch can repress the expression of the ccdB toxin protein, which can effectively kill E. coli. The gene was synthesized by GenScript and derived from a previously reported part(BBa_K5377300). Using the pUC57 plasmid as a backbone, ccdB was placed downstream of the thiM riboswitch, with the whole circuit controlled by the P_J23100.

Figure 6. OD₆₀₀ measurement under 7 mM TPP induction.

For more information, see: BBa_25ZDQ5FO

When validating the medicine-food collaboration, we employed BBa_K5377300 and conducted further characterization of it. We attempted to quantify the optimal concentration at which TPP exerts its inhibitory effect by fitting the experimental data to a four-parameter logistic (4PL) regression model.

It should be noted that IC₅₀ values may vary slightly when expressed in different vectors or in different engineered bacterial strains.

SoxR/SoxS oxidative stress response promoter enables bacteria to express anti-inflammatory factors upon sensing Hemorrhoids and High ROS environment. Based on BBa_K3771048, our experimental results demonstrate that the activity of the P_SoxS is highly dependent on ROS concentration, while exhibiting low basal expression in the absence of ROS.

For more details, you can read BBa_K3771048,

NKU-China,2024 originally designed Di-melittin(BBa_K5332002) and demonstrated its anti-inflammatory potential using a eukaryotic expression system in yeast. However, its application in prokaryotic hosts had not been explored.

This year, our team extended their work by successfully achieving prokaryotic expression of Di-melittin in E. coli. We engineered a construct with an N-terminal PelB signal peptide for secretion and a C-terminal 6×His tag for purification. Using E. coli BL21 (DE3), we confirmed expression of PelB-Di-melittin through SDS-PAGE analysis, detecting a 12 kDa band consistent with the uncleaved protein.

Our contribution demonstrates that Di-melittin can be expressed in prokaryotic systems, providing a new chassis option for future teams. This expands its potential applications, especially in engineered probiotics where prokaryotic expression is essential.

Figure 9. SDS-PAGE analysis of Di-melittin expression.

Contributions to Models

• Dataset Construction and Preprocessing: We integrated three public databases — AptaDB, Apta-Index, and the Global Nucleic Acid Aptamer Database — to collect and deduplicate data, resulting in 238 unique small molecules and 794 experimentally validated aptamer sequences. Each entry contains the molecule's SMILES structure, molecular descriptors, and the corresponding aptamer sequence. All sequences were standardized into RNA format (since aptamer function depends on secondary structure rather than nucleic acid type) to avoid noise from mixed alphabets. Byte Pair Encoding (BPE) tokenizers were separately trained for SMILES and RNA sequences. All sequences were truncated or padded to a fixed length, and the dataset was split into training, validation, and test sets at the molecule level to prevent information leakage.
• Mol2Aptamer Model Construction: We developed a conditional Transformer Encoder-Decoder architecture, where the Encoder encodes small-molecule SMILES (processed with BPE) and physicochemical features (extracted using RDKit) into a 256-dimensional conditional vector, while the Decoder autoregressively generates RNA sequences. A Conditional Variational Autoencoder (CVAE) with a latent dimension of 64 introduces sequence diversity. A Multiple Instance Learning (MIL) strategy aggregates multiple aptamer sequences per molecule to learn the objective that "at least one real sequence explains the molecule." Optimization used the AdamW optimizer (learning rate = 1e-4, weight decay = 0.01), with linear warm-up and cosine annealing scheduling. Regularization included dropout (0.1), label smoothing (ε = 0.1), and gradient clipping (norm = 1.0). The model was trained for 500 epochs with a batch size of 16. Candidate sequences were generated using greedy decoding, top-k, top-p, and temperature sampling, then filtered and reranked based on RNA secondary structure prediction (RNAfold) and physical criteria.
Figure 10. Training curves of the Mol2Aptamer across 500 epochs.
• Model Validation (Case Study: Hippuric Acid): We used mfold to predict the aptamer's secondary structure (confirming stem-loop features) and XIAO LAB/RNA Composer to construct the tertiary structure. AutoDock Vina was used to compare the known DNA aptamer for p-aminohippuric acid (from literature) and the RNA aptamer designed by Mol2Aptamer, evaluating binding score (S), RMSD (conformational similarity), and binding free energy. Using GROMACS, we built an aptamer-hippuric acid complex system (Amber99sb-ildn force field + TIP3P water model) to simulate the complex's conformational stability and analyze RMSD trajectories.
• Platform Integration: In collaboration with the PekingHSC 2025 iGEM team, the Mol2Aptamer model was integrated into the RNA-Factory platform, enabling standardized model deployment and integration with RNA structure prediction and interaction analysis tools.

• An Innovative De Novo Aptamer Design Model: We proposed Mol2Aptamer, the first model that integrates a conditional Transformer Encoder-Decoder architecture, Conditional Variational Autoencoder (CVAE), and Multiple Instance Learning (MIL) to address the one-to-many mapping between small molecules and aptamers and to enhance sequence diversity. This overcomes the limitations of traditional physics-based modeling, which relies on limited experimental data and complex recognition mechanisms.
• Development of a Practical RNA Tool Platform: In partnership with the PekingHSC 2025 iGEM team, we co-developed the RNA-Factory platform, deploying the Mol2Aptamer model in a standardized and accessible way. The platform integrates "design-verification" tools, lowering technical barriers to RNA functional element design.

• Accessible Aptamer Design: The Mol2Aptamer model integrated into the RNA-Factory platform allows other teams to directly input the SMILES and physicochemical features of a target molecule and rapidly generate high-potential aptamer candidates — without building deep learning models from scratch — saving significant development time.
• Standardized "Computation + Validation" Workflow: The demonstrated end-to-end workflow — dataset construction → model training → secondary/tertiary structure prediction → molecular docking → MD simulation — serves as a reference pipeline for other teams to verify aptamer performance. Tools such as mfold/XIAO LAB for structure prediction, AutoDock Vina for binding affinity evaluation, and GROMACS for stability analysis can be directly adopted, minimizing repeated trial and error.
• Data and Model Optimization Reference: The project provides open access to dataset information (238 molecules + 794 aptamers) and detailed hyperparameter settings (e.g., embedding dimension 256, Encoder/Decoder layers 4/6, regularization strategies), offering a foundation for other teams to replicate or optimize the model.

Our team's primary contribution lies in the development of a computational antibody directed evolution platform, designed to assist future iGEM teams in protein engineering and therapeutic antibody design. This platform integrates machine learning, structural biology, and synthetic biology principles to enable in silico optimization of antibody affinity and stability.

We trained a universal ΔΔG prediction model using the AbBind dataset, combining ESM-1b sequence embeddings with a low-rank bilinear energy model (rank = 96). This model estimates how single-point mutations affect antibody-antigen binding affinity.

$$E(H,L)= w_H^\top H + w_L^\top L + \sum_{r=1}^{R} \langle U_r, H \rangle \langle V_r, L \rangle$$

Figure 11. Predicted vs. experimental ΔΔG.

As shown in the Figure, predictions cluster around the diagonal, reflecting that the model successfully captures primary energetic trends while a few outliers remain. This filtered view highlights the model's capability to provide robust prioritization despite limited VHH-specific training data

To extend its usability to VHH and tandem VH antibodies, we introduced the heavy-light separability assumption, allowing the model to operate even in the absence of light chains.

We developed a random forest regressor trained on structural validation data to predict MolProbity Clashscore, a widely used measure of all-atom steric quality. This enables structure-based screening of candidate mutants without expensive molecular simulations.

Figure 12. Residual distribution.

Figure reveals that most errors cluster around zero, suggesting low systematic bias, with a few outliers indicating over- or under-prediction in specific cases.

By integrating ΔΔG (affinity) and Clashscore (stability) predictions, we constructed a rank-based scoring system that balances both objectives:

$$ \text{Score} = \alpha \cdot (1 - r_{\Delta\Delta G}) + \beta \cdot (1 - r_{\text{Clash}}), $$

with

$$ \alpha > \beta \, (\alpha = 2, \beta = 1) $$

Mutations with higher total scores exhibit both enhanced binding and improved structural quality. This method supports Pareto-front analysis, enabling the selection of non-dominated, interpretable candidates for experimental validation.

All scripts, including embedding extraction (generate_mutant_embeddings_esm1b_mapped.py), model training (fit_full_energy_siamese.py, train_random_forest.py), and visualization (merge_and_plot_ddg_clash.py), are modularized and publicly available. Future iGEM teams can easily adapt this framework for antibody evolution, nanobody design, or general protein optimization.

This framework provides a complete in silico alternative for teams lacking access to high-throughput experimental screening. By combining deep protein language models with low-rank bilinear learning, it enables data-driven antibody evolution entirely through computation.

A central innovation of our framework is the introduction of the heavy-light chain separability assumption, which assumes that the energetic contributions of antibody heavy and light chains can be modeled independently. This assumption is supported by the high sequence and structural homology between conventional VH domains and camelid VHH frameworks, as shown in previous studies on antibody folding and paratope energetics. Under this assumption, the model trained on VH-VL complexes can be seamlessly adapted to single-chain systems such as VHHs or tandem VH constructs by standardizing and neutralizing the light-chain input. This preserves energetic consistency while allowing the model to operate in cases where experimental data are extremely scarce.

The significance of this assumption extends beyond our own study. It offers a generalizable methodological principle for antibody and protein modeling: by identifying biophysically separable subcomponents, machine learning models can be trained on composite systems and then transferred to reduced or modified architectures. This not only expands the utility of existing antibody datasets but also demonstrates a scalable way to bridge structural biology and learning-based modeling.

Beyond the camelid anti-VEGF165 tandem VHH antibody explored here, this approach provides a reusable computational platform for rational protein design — integrating sequence embeddings, energy modeling, and structural validation — applicable to antibodies, nanobodies, enzymes, and receptors. For future iGEM teams, it exemplifies how transfer learning combined with explicit physical assumptions can make AI-driven molecular evolution both accessible and scientifically grounded, accelerating the design-build-test cycle of synthetic biology.

We constructed two key models to support an engineered bacteria-based oral drug delivery strategy. The first is a pharmacokinetic model of OMVs (outer membrane vesicles), which integrates the bacterial growth curve, an improved ADAM intestinal transport module, and a systemic circulation module to simulate the entire process of CAR peptide- and bioPROTAC-loaded OMVs from oral administration to targeting hemorrhoidal tissues. Thermodynamic analysis demonstrated that the endocytosis of OMVs across the intestinal epithelium can proceed spontaneously, providing theoretical support for their use as nanocarriers. Simulation results showed that the active proteins reached steady state in the intestinal epithelium within ~10 minutes. The steady-state concentrations of CAR peptide in external and internal hemorrhoids were 47.5 μg/ml and 40 μg/ml, respectively, while those of bioPROTAC were 21.5 μg/ml and 18 μg/ml. These results suggest that the oral delivery system holds therapeutic potential for external, internal, and mixed hemorrhoids.

Figure 13. The results of OMV pharmacokinetics.

The second model is a "Expiry Date Circuit", designed to regulate the life cycle of engineered bacteria via a riboswitch and a small-molecule ligand. We employed ordinary differential equations and Hill functions to describe the inhibition and recovery of the toxin protein CcdB expression by the ligand. Our analysis revealed that, even under extremely low ligand doses (1 nM) or very high half-inhibitory concentrations, the time for CcdB to reach the lethal threshold remained stable at around 42 hours. This indicates that patients would only need to supplement the ligand approximately every 41 hours, demonstrating the system's robustness and controllability. The circuit thus provides theoretical support and a design framework for the safe clearance and timed therapeutic action of engineered bacteria.

Figure 14. The results of expiry-date circuit ODE.

Our work provides valuable references for other iGEM teams in the following aspects: the modeling framework of OMV-mediated drug delivery, gene circuit dynamics modeling methods, thermodynamic analysis of transmembrane processes, and multi-scale integration of physiological parameters. These results not only validate the feasibility and effectiveness of oral engineered bacterial systems but also offer reusable modeling strategies and theoretical tools for future teams developing bacteria-based carriers or controllable genetic circuits for therapeutic applications.

This contribution provides a comprehensive, optimized computational biology work flow and automated script designed to assist iGEM teams in accurately and efficiently calculating the binding free energy of protein-ligand/peptide complexes, specifically employing the Sliding Molecular Dynamics (SMD) and Umbrella Sampling (US) methods. This toolkit resolves critical challenges commonly encountered in MD simulations, such as complex system orientation, precise simulation box sizing, and automation of multi-step equilibration processes, significantly lowering the barrier for iGEM teams to conduct biological research using computational methods.

You can read details at Model.

Comtributions to IHP

To align Human Practices with the engineering principles of iGEM, we proposed the Engineering IHP Framework, a cyclic model integrating investigation, practice, feedback, and integration into a coherent iterative process.

Throughout our entire IHP work, this framework served as the backbone of our practice, guiding every activity from problem identification to solution refinement. It transformed HP from a linear workflow into a dynamic feedback system, ensuring that social insights and experimental design evolved together.

By applying this framework across all stages, we strengthened the logic and rigor of our HP process. We share it here as a reproducible tool that future teams can adapt to structure their own HP and demonstrate how feedback drives responsible improvement.

During our IHP and engineering design, we developed the concept of Medicine-Food Collaboration to address a critical challenge in living therapeutics: how to safely stop the treatment once the therapeutic bacteria are active in the body.

We proposed that engineered probiotics could respond to specific dietary cues to sustain or terminate their function. In this design, patients can control treatment simply by adjusting their diet—for instance, by pausing the intake of certain foods for several days.

Through multiple iterations, we refined this concept in response to expert advice and practical limitations, such as food selection difficulty and patient compliance. The final version integrates biological control with daily behavior, offering a novel and patient-friendly regulation model for future synthetic biology therapeutics.

To share this innovation, we compiled the Medicine-Food Collaboration Handbook, summarizing our theoretical basis, design logic, and practical insights for all iGEM teams to reference, learn from, and further develop.

You can see detailed iteration process on the Engineering and IHP pages.

At the very beginning of topic selection, we noticed the potential stigma surrounding hemorrhoids—a condition often associated with shame and limited public attention.

During our Mother's Day education activity, we focused on pregnant women, a group vulnerable to both hemorrhoids and mental health issues. Questionnaire results revealed a key insight: external stigma outweighed self-stigma, shaping patients' willingness to seek help.

Further interviews with doctors and patients confirmed that stigma and embarrassment often cause treatment delays, with many individuals only seeking medical care at severe stages.

Recognizing that most people learn about hemorrhoids online, we conducted a systematic social media analysis in China, combining qualitative exploration of stigma formation pathways with a machine learning-based classification model to expand the scope of stigma research.

Our findings were integrated into Principle III: Patient-Centered Approach within our IHP framework and later informed our public education materials.

We hope our work can inspire future iGEM teams to investigate disease-related stigma—whether in infectious, chronic, or mental disorders—and collectively contribute to a more understanding and destigmatized healthcare environment.

To promote destigmatization and spread awareness of synthetic biology, we carried out a series of educational and outreach activities targeting different audiences.Here, we share several open-access resources that other iGEM teams can freely reference and adapt in their own projects:

Figure 16.Our Mother's Day education activity poster