AI Peptides for Influenza HA Binding

ISEF Category: Microbiology

Subcategory: Virology  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Influenza changes fast, but the stalk of hemagglutinin, or HA, stays more conserved than the rest of the virus. That makes it a tempting target for a mini-binder, a small peptide built to stick to one spot. You can use AI design tools to make and rank candidates before any wet lab work. That gives you a real shot at original research, even if you do not start with a bench lab.

What Is It?

This project asks you to design tiny protein-like binders that may attach to the HA stalk on influenza. Think of HA as a key on the virus surface. The stalk is the part near the base of that key. It changes more slowly than many other viral parts, so a binder that fits there could block infection more reliably than one aimed at a rapidly mutating site.

You are not making a drug in one step. You are building a short list of designs with computational tools. RFdiffusion helps propose backbone shapes, ProteinMPNN helps choose amino acids that fit that shape, and AlphaFold-Multimer estimates whether your binder and HA might form a stable complex. pAE, or predicted aligned error, gives you one way to judge how confidently the model thinks the two proteins fit together. Lower pAE often suggests a better-predicted interface.

Why This Is a Good Topic

This is a strong science fair topic because you can test many design choices without needing a full virology lab. You can compare candidate binders, scoring rules, and target regions, then ask which choices produce the best predicted binding. The project connects to influenza prevention, antiviral design, and protein engineering. You can also learn how to read structural models, compare rankings, and explain why one design looks better than another.

Research Questions

• How does binder length affect AlphaFold-Multimer pAE for influenza HA stalk targets?
• What is the effect of targeting different conserved HA stalk residues on predicted interface confidence?
• Does RFdiffusion-generated backbone shape improve pAE compared with random peptide scaffolds?
• To what extent do ProteinMPNN sequence choices change predicted stability for the same HA-bound backbone?
• Which top-ranked designs stay stable across repeated model runs with different random seeds?
• How does adding a short helix-rich scaffold affect predicted contact density at the HA interface?
• What is the effect of using HA sequences from different influenza strains on binder ranking?

Basic Materials

• Laptop or desktop computer with internet access.
• Google account for Colab access.
• Python installed locally or used through Colab notebooks.
• Public HA structure files from the Protein Data Bank.
• FASTA sequences for influenza HA stalk regions.
• Spreadsheet software for tracking design scores.
• Screen capture tool for saving model outputs.
• Notebook for recording model settings and ranking rules.

Advanced Materials

• Access to a university or collaborator compute server.
• GPU-enabled workstation or cloud notebook with enough memory for structure prediction runs.
• Local installation of Python, PyMOL, and supporting bioinformatics packages.
• Protein Data Bank structure files for multiple HA subtypes.
• Multiple influenza HA sequence alignments from public databases.
• Scripts for batch scoring, ranking, and filtering designs.
• Optional molecular visualization software for interface inspection.
• A documented pipeline for AlphaFold-Multimer runs with repeated seeds.

Software & Tools

• Google Colab: Runs shared notebooks for RFdiffusion and ProteinMPNN without local setup.
• Python: Organizes design batches, scores, and ranking tables.
• AlphaFold-Multimer: Predicts whether your peptide and HA form a plausible complex.
• PyMOL: Lets you inspect where the binder sits on the HA stalk.
• ImageJ: Can help if you compare saved model images or annotate figures for a poster.

Experiment Steps

1. Define the design target by choosing one HA stalk structure and one clear binding region.
2. Set your comparison plan by deciding which design variable you will change first, such as scaffold shape, sequence, or target subtype.
3. Build a scoring rubric that ranks candidates using pAE, interface contacts, and simple stability checks.
4. Run repeated predictions so you can see which designs stay high-ranking across random seeds.
5. Filter the output to a small top-3 set and document why each one beats the others.
6. Plan a validation handoff by writing the exact sequence and structure data a partner lab would need for cell-culture follow-up.

Common Pitfalls

• Choosing an HA target with a highly variable loop instead of the conserved stalk, which makes the design less useful.
• Treating a single AlphaFold-Multimer run as proof, which ignores run-to-run variation.
• Ranking only by pAE, which can miss bad interface geometry or poor overall shape.
• Mixing HA structures from different strains without tracking the sequence differences, which makes comparisons messy.
• Ignoring whether the binder sequence is chemically realistic, which can produce designs that look good on paper but are hard to test later.

What Makes This Competitive

A competitive project would go beyond making one or two peptide designs. You would compare several target regions, several scoring rules, and several random seeds, then defend your final top-3 with clear evidence. Strong entries also explain why the chosen designs matter for influenza escape, not just why they score well in one model. If you can show a repeatable pipeline and a careful analysis of failure cases, your project starts to look like real research.

Project Variations

• Compare binder design against HA stalks from different influenza A subtypes to test whether one scaffold generalizes better.
• Swap the scoring focus from pAE to interface contact count and see whether the ranking changes.
• Design a slightly longer peptide versus a 60-aa mini-binder to test whether added length improves predicted stability.

Learn More

• Protein Data Bank: Search for influenza hemagglutinin structures and download coordinate files for modeling.
• PubMed: Search for review articles on influenza HA stalk antibodies, peptide binders, and structure-based antiviral design.
• NIH National Library of Medicine Bookshelf: Find free background chapters on protein structure, binding, and influenza biology.
• MIT OpenCourseWare: Look for free materials on structural biology, bioinformatics, and protein engineering.
• NCBI Virus: Find influenza sequence data and strain comparisons for target selection.
• RCSB PDB Learning Resources: Read short guides on interpreting protein structures and interfaces.