To view the code of model construction
please click
https://gitlab.igem.org/2025/smu-union-china
to visit our code file.
Overview

Our integrated modeling framework links molecular-level protein modeling, microbial growth kinetics, and population-level infectious disease dynamics to evaluate and optimize antibody-based prevention strategies against influenza. The framework has four complementary aims:
A. Protein Modeling: predicting the structural binding of Medi8852 scFv to H1N1 hemagglutinin (HA).
B. Growth Curve Simulation:quantitatively model engineered bacterial growth to evaluate the metabolic burden imposed by plasmid expression systems.
C. Infectious Disease Dynamics: Simulating the impact of antibody interventions on influenza transmission using the SEIAR model.
D. Synthetic Biology Component:Evaluating an RNA thermosensor system as a potential regulatory element.
Protein Modeling
Protein Structure Prediction
Homology Modeling
To predict the binding efficiency of our Medi8852 scFv sequence against the HA molecule of H1N1, we aimed to model and assess the binding complex through protein modeling techniques. Initially, we procured the PDB structure of H1N1 (PDB: 1RUZ) and proceeded to construct a homology model of Medi8852 scFv using Swiss-Model.
The structure of Medi8852 scFv was built based on the template X-ray diffraction structure of the MEDI8852 Fab Fragment (PDB: 5JW5), as depicted in Figure 1. Following the model construction, a quality assessment of the Medi8852 structure was conducted using the PROCHECK module of the SAVES v6.1 server, as shown in Figure 2.
The model demonstrated good stereochemical geometry, with no residues falling into disallowed regions and an overall acceptable G-factor. Consequently, it was deemed suitable for subsequent docking and simulation steps.

Structural validation was performed using PROCHECK, and the results are summarized below:

Overall, these results indicate that the selected complex had good stereochemical quality, with no residues falling in the disallowed region and only minimal deviations detected.
Molecular Docking of MEDI8852 scFv with H1N1 Hemagglutinin
Pre-docking:Protein Cleaning
The pre-docking phase involved preparing both the receptor (H1N1) and the ligand (Medi8852 scFv) for molecular docking. Using AutoDock, the PDB file of H1N1 (PDB: 1RUZ) was imported, water molecules were removed, hydrogens were added, and the structure was prepared as the receptor. It was then exported as a PDBQT file, completing the receptor preparation. Similarly, the PDB file of Medi8852 (PDB: 5JW5) was imported, hydrogens were added, the structure was prepared as the ligand, torsion bonds were identified, and the file was exported as a PDBQT file. This process completed the preparation (cleaning) of both docking molecules.
AutoDock docking
The initial attempt to perform docking using AutoDock failed due to the large molecular weight of the ligand. Large ligands, with their high flexibility and numerous rotatable bonds, significantly increase computational complexity, often exceeding AutoDock's capabilities for conformational sampling. Consequently, AutoDock was unable to complete the docking process successfully. To address this, we switched to ZDOCK, an alternative docking tool that provides better support for large molecules.
ZDOCK-Rigid-body Docking
ZDOCK 3.0.2 was employed for the initial rigid-body docking of the MEDI8852 scFv and the H1N1 HA protein. The cleaned structures (as prepared in AutoDock,) were used as input for ZDOCK.
Parameters:
Rotational sampling: 6°
Translational sampling: 1.2 Å
Number of poses generated: 2,000
Scoring function: ZRank
Top clusters retained: 10
The best pose from the top-ranked cluster had a ZRank score of −1308.4 and an interface area of 1,584 Ų. This pose was selected for subsequent molecular dynamics simulations.
Previous studies have demonstrated that the MEDI8852 scFv binds to the stem region of the H1N1 HA protein. Consistent with these findings, we examined the docking interface in PyMOL and confirmed that the predicted binding site was located in the HA stem. Among the generated docking poses, the Top 1 predicted complex was selected for further quality assessment.
PyMOL visualization
PyMOL visualization confirmed that the MEDI8852 scFv binds to the stem region of HA (cyan), consistent with previous studies. The left panel shows the overall docking orientation, while the right panel highlights the interaction interface. Several hydrogen bonds and hydrophobic contacts were observed, with residues such as TYR659 and ASN628 involved in stabilizing the binding. This stem-targeting binding mode is significant because it prevents the conformational rearrangements of HA necessary for membrane fusion, thereby neutralizing the virus.

iMODS - Flexibility Assessment
To assess the flexibility and stability of the scFv–HA complex, the structure was analyzed using the iMODS online server (job ID 0600205804037). The analysis included normal mode analysis (NMA) and elastic network modeling (ENM).
The iMODS analysis confirms that the scFv–HA complex is flexible enough to adapt to binding-induced conformational changes while maintaining overall stability. This validated structure serves as a reliable input for further molecular dynamics simulations and epidemiological modeling.
Binding Affinity Prediction
To further evaluate the interaction strength of the docked complex, we used PRODIGY after removing small molecules and retaining only the two protein chains.These results indicate that the MEDI8852 scFv–HA stem complex exhibits a predicted binding free energy of –11.4 kcal/mol and a nanomolar dissociation constant (Kd = 6.8 nM), suggesting a high-affinity interaction under 32℃. The interface is stabilized by a combination of charged–hydrophobic and hydrophobic–hydrophobic contacts, which together account for over 60% of the binding interactions.

Antibody-Antigen Binding Model Construction
This simulation is based on the Langmuir binding model, which predicts the relationship between antigen concentration and the binding ratio (θ) under varying antibody concentrations.
Core Equation

Assumptions and Inputs
Biological Significance
Key Findings
Clinical Context

Growth Curve Simulation
Modeling the Growth Dynamics of Lactobacillus rhamnosus GG
To accurately characterize the growth dynamics of bacteria, we fitted the experimental OD measurements over time to three classical growth models: Logistic, Gompertz, and Richards. By comparing the goodness-of-fit (R²) across different datasets, the Gompertz model was found to provide the best fit. Based on this model, we extended the growth curve prediction from the original 25 days to 35 days, successfully forecasting the future growth trend of the bacteria. This provides a quantitative basis for further experimental design and microbial population management.
Growth Curve Analysis of Plasmid-Transformed Lgg
To compare the growth burden imposed by different plasmids on bacterial hosts and investigate whether plasmid loss during liquid culture is related to antibiotic selection pressure, we analyzed bacterial growth curves.
First, the growth curve of the wild-type LGG strain (dark red) demonstrates continuous and efficient growth across all figures, ultimately reaching a high OD value of approximately 3.0. This indicates unimpaired metabolic activity and the absence of any plasmid-related burden.
For engineered strains carrying plasmids, taking the first figure as an example, the growth curve of PG4 without erythromycin (blue) shows rapid initial growth that later stagnates, with the final OD value stabilizing around 1.7. The growth pattern of PG4ery with erythromycin (purple) is similar to PG4, reaching a comparable final OD value. In the second figure, RNA1-3 without erythromycin (orange) and RNA1-3ery with erythromycin (green), as well as TIpA without erythromycin (yellow) and TIpAery with erythromycin (light blue) in the third figure, all plasmid-carrying engineered strains - regardless of erythromycin supplementation - exhibit significantly lower final OD values than the wild-type LGG. This clearly demonstrates that plasmid introduction imposes a substantial growth burden on the bacteria, consuming metabolic resources and limiting growth potential.
Regarding the relationship between plasmid loss and antibiotic pressure, the presence or absence of erythromycin differentially affects the growth of engineered strains. Although all erythromycin-treated strains show growth limitation due to plasmid maintenance, variations exist compared to their untreated counterparts. For instance, in the first figure, PG4ery and PG4 show similar final OD values; in the second figure, RNA1-3ery exhibits a slightly higher final OD than RNA1-3; while in the third figure, TIpAery achieves a considerably higher final OD than TIpA. This suggests that erythromycin, as a selection antibiotic, may influence plasmid stability. During liquid culture, antibiotic presence likely reduces plasmid loss by exerting selective pressure against plasmid-free (or plasmid-lost) cells, thereby promoting the growth of plasmid-retaining bacteria. This indirectly indicates that plasmid loss during shaking culture is associated with antibiotic-mediated selection pressure. Through this selective mechanism, antibiotics affect plasmid retention, consequently influencing bacterial growth performance.
Strain | Inflection Point (t₀) | Growth Rate (r) | Carrying Capacity (K) |
---|---|---|---|
PG4 | 5.2373 | 0.2279 | 1.7038 |
RNA1-3 | 5.0999 | 0.2295 | 1.7374 |
TlpA | 6.0150 | 0.2382 | 1.7353 |
PG4ery | 9.3405 | 0.1957 | 1.6814 |
RNA1-3ery | 9.7354 | 0.1240 | 2.3058 |
TlpAery | 11.8860 | 0.1325 | 2.7020 |
LGG (WT) | 11.0598 | 0.1309 | 3.0116 |
Validation of Plasmid Retention Analysis
To evaluate the stability of our engineered constructs, colony-forming units (CFUs) were measured on both selective and non-selective plates. The resulting plasmid retention rates showed clear differences among strains. At 12 hours, only 5.9% of PG4 cells retained the plasmid, RNA1-3 exhibited an even lower retention of 1.7%, while TlpA maintained the highest rate at 2.5%. These results indicate substantial plasmid loss during early growth, with stability varying depending on the construct.
Table. Plasmid Retention Rates of Different Strains at 12h
Strain | CFU/ml on Selective Plate | CFU/ml on Non-Selective Plate | Plasmid Retention Rate (Selective/Non-Selective) |
---|---|---|---|
PG4-12h | 4.6 × 107 | 7.8 × 108 | 5.9% |
RNA1-3-12h | 1.57 × 107 | 9.3 × 108 | 1.7% |
TlpA-12h | 1.55 × 107 | 6.2 × 108 | 2.5% |
Despite the low plasmid retention observed in CFU counts, growth curve analysis suggested that the engineered strains did not experience severe growth defects. This apparent discrepancy may be due to insufficient selective pressure to maintain plasmids or compensation by plasmid-free cells dominating the population over time.
To capture plasmid dynamics more accurately, we propose incorporating 6–12 h time-series measurements of CFU and plasmid copy number, alongside simultaneous monitoring of culture pH. This allows modeling of both plasmid loss and the environmental effects on antibiotic activity. Once these data are parameterized into the extended model, different optimization strategies can be tested in silico. For example,extending antibiotic half-life by using pH-buffered media or more stable antibiotics and reducing plasmid loss through toxin–antitoxin stabilization systems.
Infectious Disease Dynamics SEIAR model
Fitting of Real-World Data for an Influenza A Outbreak
Model Construction
We employ the SEIAR model to simulate influenza transmission, which classifies the population into five compartments: (Used the number of people with influenza A in the United States from the 40th week of 2024 to the 28th week of 2025)
S (Susceptible): Individuals at risk of infection
E (Exposed): Infected individuals in the latent period
I (Infectious): Symptomatic infected individuals
A (Asymptomatic): Asymptomatic infected individuals
R (Recovered): Recovered individuals
Fitting Equation of SEIAR
SEIAR Model Fitting Equations (System of Differential Equations):

The formula for the basic reproduction number R0:

Fitted Parameters:
β (Symptomatic transmission rate): 1.0
k (Relative transmission coefficient for asymptomatic cases): 0.5
Fixed Parameters: ω=0.53, p=0.14, γ=0.23, γ₁=0.24

Model Optimization
An iterative optimization strategy is adopted to estimate parameters β and k:
Rationale:
1. Objective Function: Minimize the mean squared error (MSE) between the modelpredicted (I+A) values and the actual data.
2. Optimization Algorithm: LBFGSB (a quasiNewton method with bound constraints).
3. Iterative Strategy:
(1) Initial parameter ranges: β ∈ [0.2, 1.5], k ∈ [0.1, 0.8].
(2) Perform 3 iterations, narrowing the search range by 20% in each iteration.
(3) Update the central parameter values after each iteration.
4. Final Selection: Choose the parameter combination with the smallest MSE from all iteration results.


Rationality Analysis of SEIAR Parameters
1. Optimized Parameters (beta, k)beta (Symptomatic Transmission Rate): 1.1737
Rationale:
Within the reference range of 0.5–1.5, consistent with the transmission characteristics of influenza.
The model's MSE = 114534.30 indicates a good fit, supporting the validity of this value.
k (Relative Transmission Coefficient for Asymptomatic Cases): 0.1211
Rationale:
Within the reference range of 0.1–0.8. The value < 1 aligns with the principle that asymptomatic transmission is weaker than symptomatic transmission. When combined with beta, it accurately reflects realworld data trends, demonstrating logical consistency.
2. Fixed Parameters
ω (Inverse of Incubation Period): 0.53
Rationale:
Corresponds to an incubation period of 1/omega ≈ 1.9 days, within the typical influenza incubation range (1–3 days).
p (Proportion of Asymptomatic Cases): 0.14
Rationale:
14% falls within the documented range for influenza asymptomatic infections (10%–40%), matching epidemiological findings.
γ (Symptomatic Recovery Rate): 0.23
Rationale:
Corresponds to a disease duration of 1/γ ≈ 4.3 days, consistent with the 3–7day recovery period for symptomatic influenza cases.γ1 (Asymptomatic Recovery Rate): 0.24
Rationale:
Corresponds to a disease duration of 1/γ1 ≈ 4.2 days, slightly shorter than symptomatic cases, which aligns with the shorter duration of asymptomatic infections.
3.Overall Rationale Summary
Parameters are biologically reasonable and yield accurate epidemic fitting.
Common expressions of other parameters in infectious disease models

The Impact of Early Use of Medi8852 scFv and 1G01 scFv on the Transmission of Influenza A
According to published studies, MEDI8852 effectively blocks early infection transmission by reducing the infection rate among contacts. In ferret models of H1N1pdm09 infection, prophylactic intraperitoneal administration of MEDI8852 to uninfected contact ferrets prevented airborne transmission in most cases: 3 out of 4 contact ferrets became infected, corresponding to approximately 75% transmission blockage. Donor ferrets still exhibited high viral titers, and contact ferrets showed viral shedding for 5–11 days with a peak on day 3 post-infection, indicating that the effect primarily prevents new infections rather than reducing viral load in infected donors.
This empirical value is strongly supported by our in silico predictions. We calculated the binding affinity (Kd) of the MEDI8852 scFv-HA complex to be 6.8 nM at 32°C, approximating upper respiratory tract conditions. This high-affinity interaction suggests a potent neutralizing capacity, with the Langmuir model predicting a theoretical in vitro neutralization efficiency (η) exceeding 90% at clinically relevant antibody concentrations (50-300 nM). This agreement—from in silico affinity to in vivo efficacy to epidemiological modeling—provides a coherent and validated foundation for the 75% transmission reduction parameter.
In SEIAR model terms, MEDI8852 reduces the transmission rate (β) and the relative infectiousness of asymptomatic individuals (k), with adjusted values:
β_new = β_old × (1 - Efficacy × Coverage)。
At 30% population coverage → β_new ≈ 0.899, k_new ≈ 0.0939
At 50% coverage → β_new ≈ 0.739, k_new ≈ 0.0757
1G01, an anti-neuraminidase monoclonal antibody, reduces viral loads and accelerates recovery in both donors and recipients, as shown in guinea pig models. Administration of 1G01 to donor guinea pigs shortened recovery from 10 days (control) to 6 days, and in recipient guinea pigs from 10 days to 8 days, reducing potential transmission opportunities. In the SEIAR model, this effect is represented by increased recovery rates.
γ_new = γ_old / (1 - the proportion of duration reduction)
γ_new ≈ 0.329,
γ₁_new ≈ 0.343
Assumptions for Antibody Intervention in SEIAR Model
A. Timing and efficacy of MEDI8852: Prophylactic or early administration (within 1–2 days post-infection) achieves approximately 75% transmission blockage.
B. Population coverage: 30–50% of the population receives the antibody.
C. 1G01 effect on infection duration: Timely administration to both donors and recipients shortens the infectious period by ~30% (range: 20–40%).
D. Combined use of MEDI8852 and 1G01 may both decrease the probability of contact infection and shorten the duration of infection, achieving a synergistic protective effect in the population.
E. ScFv nasal administration equivalence: Nasal delivery of scFv antibodies is assumed to have equivalent efficacy to intraperitoneal monoclonal antibody administration.
Parameter | Original | Intervention (30–50% Coverage) |
---|---|---|
β | 1.1743 | 0.899 – 0.739 |
k | 0.1211 | 0.0939 – 0.0757 |
γ | 0.23 | 0.329 |
γ₁ | 0.24 | 0.343 |
ω | 0.53 | - |
p | 0.14 | - |
No intervention Scenario

Vaccination coverage in the population under randomized transmission
RANDOM1

RANDOM2

RANDOM3

Original R₀ is approximately 4.47 (calculated by substituting parameters), which falls within the R₀ range (1–6) of most infectious diseases, indicating a moderate transmission intensity. The distribution of post-intervention R₀ decreases due to the increase in vaccination rate and recovery rate, reflecting the core logic that "intervention measures can effectively control the epidemic," which is consistent with the actual epidemic control outcomes.

RNA thermosensor system
Based on the ROSE-like RNA thermometer sequence
CCGGGCGCCCUUCGGGGGCCCGGCGGAGACGGGCGCCGGAGGUGUCCGACGCCUGCUCGUCCAGUCUUUGCUCAGUGGAGGAUUACUAGAUGAGGA, the command RNAfold --temp 37 < rose.fa > rose_37.fold was used to predict changes in the RNA secondary structure at different temperatures by varying the temperature parameter "temp". At 37°C, the predicted RNA exhibits a specific folding pattern (as shown in the first figure), while when the temperature increases to 50°C, its structure undergoes significant changes (as shown in the second figure).
However, this initial prediction significantly diverged from our experimental observations. The experiments revealed a distinct "fluorescence dip" in gene expression controlled by this RNA thermometer around 30°C, strongly suggesting that key local structural rearrangements occur well below the 50°C range. This discrepancy prompted a deeper reflection on our predictive model, revealing two core issues:
1. Sequence Specificity Was Overlooked:Our initial modeling directly applied the consensus sequence derived from Gram-negative bacterial ROSE elements. However, the 5'-UTR used in our system, originating from Lactobacillus rhamnosus, has a significantly higher GC content (69%) compared to typical E. coli sequences (53%), and the ribosome binding site spacing differs by 2 nucleotides. These factors collectively caused the initial model to overestimate the global thermal stability of the structure, thereby mispredicting its melting temperature.
2. Limitations of Static Folding:Tools like RNAfold predict the thermodynamically most stable structure but fail to capture kinetic intermediates or local conformational fluctuations that might be crucial for translation initiation at physiological temperatures. The observed "first decrease, then increase" pattern likely stems from the temporary formation of a metastable "trapping" structure within the 25–30°C window that blocks ribosome access—a structure overlooked by standard free energy minimization calculations.
Therefore, the initial computation was not incorrect, but rather oversimplified. It accurately identified the final thermal melting point (50°C) but failed to reveal the fine structural changes relevant to functional regulation occurring at lower temperatures. The experimental data served as the key to uncovering these functionally relevant states masked by the global free energy minimum.
Discussion on Future Work
To address the current challenges, we have formulated a systematic, data-driven research plan to elucidate the regulatory mechanism and optimize system performance:
1. Constructing a High-Resolution Four-Dimensional Dataset: The core of future work is to establish a high-density, high-precision dataset encompassing "Temperature–Time–Fluorescence Intensity–mRNA Half-life". By intensive sampling at 2°C intervals between 25–42°C, with increased biological replicates and time points, we will obtain reliable data sufficient for complex kinetic modeling, fundamentally reducing the risk of overfitting.
2. Developing a Physiologically Relevant Hybrid Computational Model: We will move beyond single static structure prediction towards a more advanced hybrid model combining ensemble folding kinetics and Monte Carlo simulations.
Constraints:The model will incorporate Lactobacillus rhamnosus-specific ribosomal binding free energy as a key constraint.
Reverse Optimization: We will use the experimentally observed "25-30°C fluorescence dip" as a hard boundary condition for the model, to inversely optimize parameters such as free energy (ΔG) and translation initiation rate constant (k_init). Our goal is to make the model's output precisely recapitulate this atypical temperature response curve, compressing the prediction error to within ±1°C.


In summary, FluBlockers is a novel, intelligent, and safe probiotic-based strategy against Influenza A.
It can:


References
[1]Kallewaard NL, Corti D, Collins PJ, Neu U, McAuliffe JM, Benjamin E, Wachter-Rosati L, Palmer-Hill FJ, Yuan AQ, Walker PA, Vorlaender MK, Bianchi S, Guarino B, De Marco A, Vanzetta F, Agatic G, Foglierini M, Pinna D, Fernandez-Rodriguez B, Fruehwirth A, Silacci C, Ogrodowicz RW, Martin SR, Sallusto F, Suzich JA, Lanzavecchia A, Zhu Q, Gamblin SJ, Skehel JJ. Structure and Function Analysis of an Antibody Recognizing All Influenza A Subtypes. Cell. 2016 Jul 28;166(3):596-608.
[2]Waterhouse A, Bertoni M, Bienert S, Studer G, Tauriello G, Gumienny R, Heer FT, de Beer TAP, Rempfer C, Bordoli L, Lepore R, Schwede T SWISS-MODEL: homology modelling of protein structures and complexes. Nucleic Acids Res 46, W296-W303. (2018) PubMed logo29788355 DOI logo10.1093/nar/gky427
[3]Pierce BG, Hourai Y, Weng Z. (2011) Accelerating Protein Docking in ZDOCK Using an Advanced 3D Convolution Library. PLoS One 6(9): e24657
[4]Schrödinger, LLC. (2015). The PyMOL Molecular Graphics System, Version 1.8. November.
[5]Vangone A. and Bonvin A.M.J.J. "Contact-based prediction of binding affinity in protein-protein complexes", eLife, 4, e07454 (2015).
[6]Paules CI, Lakdawala S, McAuliffe JM, Paskel M, Vogel L, Kallewaard NL, Zhu Q, Subbarao K. The Hemagglutinin A Stem Antibody MEDI8852 Prevents and Controls Disease and Limits Transmission of Pandemic Influenza Viruses. J Infect Dis. 2017 Aug 1;216(3):356-365. doi: 10.1093/infdis/jix292. PMID: 28633457; PMCID: PMC5853468.
[7]Tan J, O'Dell G, Hernandez MM, Sordillo EM, Kahn Z, Kriti D, van Bakel H, Ellebedy AH, Wilson PC, Simon V, Krammer F, McMahon M. Human Anti-neuraminidase Antibodies Reduce Airborne Transmission of Clinical Influenza Virus Isolates in the Guinea Pig Model. J Virol. 2022 Jan 26;96(2):e0142121. doi: 10.1128/JVI.01421-21. Epub 2021 Oct 20. PMID: 34669506; PMCID: PMC8791283.
[8]Plotnik D, Sager JE, Aryal M, Fanget MC, Peter A, Schmid MA, Cebrik D, Mogalian E, Boundy K, Yeh WW, Griffin P, Reyes M. A phase 1 study in healthy volunteers to investigate the safety, tolerability, and pharmacokinetics of VIR-2482: a monoclonal antibody for the prevention of severe influenza A illness. Antimicrob Agents Chemother. 2024 Apr 3;68(4):e0127323. doi: 10.1128/aac.01273-23. Epub 2024 Feb 20. PMID: 38376227; PMCID: PMC10988998.
[9]Noh JY, Song JY, Hwang SY, Choi WS, Heo JY, Cheong HJ, Kim WJ. Viral load dynamics in adult patients with A(H1N1)pdm09 influenza. Epidemiol Infect. 2014 Apr;142(4):753-8. doi: 10.1017/S0950268813001672. Epub 2013 Jul 18. PMID: 23866846; PMCID: PMC9151081.