IL-6 Binder Design With Generative AI

ISEF Category: Biomedical Engineering

Subcategory: Synthetic Biology  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A tiny protein can act like a custom lockpick for a disease signal. If you can design one that fits IL-6, you are working on the same kind of molecular targeting used in real drug discovery. That makes this project feel like AI plus biology, but it still comes down to good controls and careful scoring. You do not need to discover a new machine learning model to make this impressive.

What Is It?

IL-6 is a signaling protein, also called a cytokine. Your body uses cytokines like text messages between cells. In cytokine storms, those messages get too loud and too frequent, which can drive dangerous inflammation. A binder is a molecule that sticks to IL-6 and may block its signal. Your project asks whether a computer can design a small protein that fits IL-6 well enough to be worth deeper testing.

RFdiffusion helps generate new protein shapes, and ProteinMPNN helps choose amino acid sequences that can fold into those shapes. Think of RFdiffusion as sketching a key shape and ProteinMPNN as choosing the teeth on that key. Then you score the design for developability. Solubility asks whether the protein may stay dissolved. Aggregation asks whether it may clump. Immunogenicity asks whether the immune system may treat it as a threat. Those extra checks matter because a design that binds well on screen can still fail in a real body.

Why This Is a Good Topic

This topic works well for a science fair because you can change one design choice at a time and measure the effect with clear scoring metrics. It connects to a real medical problem, since IL-6 plays a major role in severe inflammation. You can also compare designs, rank them, and defend why one looks better than another. That gives you real research, not just a one-off computer run.

Research Questions

• How does the choice of RFdiffusion seed affect predicted binding score to IL-6?
• What is the effect of different ProteinMPNN sequence samples on predicted solubility?
• Does adding a developability filter change which IL-6 binder designs survive ranking?
• To what extent do interface contact counts predict overall binder ranking for IL-6?
• Which design features best reduce predicted aggregation while keeping strong IL-6 binding?
• How does immunogenicity screening change the top candidate list for IL-6 binders?

Basic Materials

• Laptop or desktop computer with enough memory to run Colab notebooks.
• Google account for free Colab access.
• Web browser with stable internet access.
• Open-source protein visualization tool such as PyMOL Educational or UCSF ChimeraX.
• Spreadsheet software for ranking design scores.
• Free access to Protein Data Bank structures for IL-6 or IL-6 complex templates.
• PubMed access for reading background papers on IL-6 and cytokine storms.

Advanced Materials

• Access to a university workstation or cloud GPU for repeated design runs.
• Local installation or advanced access to PyMOL, ChimeraX, and structure-analysis scripts.
• Access to academic NetMHCpan use or an approved institutional interface.
• SoluProt or a comparable solubility prediction workflow.
• Molecular docking or interface-analysis software for cross-checking design rankings.
• Sequence analysis scripts in Python or R for batch scoring and plotting.
• Protein structure datasets for comparison against known cytokine binders.

Software & Tools

• Google Colab: Runs RFdiffusion and ProteinMPNN notebooks without local GPU hardware.
• PyMOL: Lets you inspect the binder interface and compare candidate structures.
• UCSF ChimeraX: Helps you visualize contacts, surfaces, and shape complementarity.
• Python: Automates ranking, plotting, and summary statistics for many designs.
• PubMed: Lets you search review articles and primary papers on IL-6, cytokines, and protein design.

Experiment Steps

1. Define the exact IL-6 binding problem you want to solve, including the target structure and the interface region you will score.
2. Choose one design strategy first, then keep the other settings fixed so you can compare outputs fairly.
3. Set up a ranking plan that combines binding prediction with developability filters, not binding alone.
4. Build a comparison set so you can judge your designs against baseline sequences or published binders.
5. Plan how you will inspect structural features, such as interface size, surface charge, and exposed hydrophobic patches.
6. Decide which statistical test or ranking method will separate promising designs from weak ones.

Common Pitfalls

• Ranking only by predicted binding, which can hide sequences that are likely to aggregate or express poorly.
• Comparing designs made with different target settings, which makes the results hard to interpret.
• Trusting a single AI-generated sequence without checking whether the interface geometry makes physical sense.
• Ignoring developability scores, which can leave you with a binder that looks good on screen but fails basic protein quality checks.
• Using too few comparison designs, which makes it hard to tell whether your result beats a random baseline.

What Makes This Competitive

A stronger project would not just produce one binder candidate. It would compare several design pipelines, use clear baselines, and explain why the top model wins. You can also get more credit by testing whether developability filters change the ranking in a meaningful way. The best version shows that you understand both protein design and the limits of the scoring tools.

Project Variations

• Design binders for a different cytokine target, such as IL-1 or TNF-alpha, and compare how target shape changes design success.
• Swap the analysis focus to developability only, then rank a library of candidate sequences by solubility, aggregation risk, and immunogenicity.
• Compare miniprotein binders against peptide binders or antibody fragments using the same scoring framework.

Learn More

• NIH PubMed: Search review articles on IL-6, cytokine storms, and protein therapeutics to build background knowledge.
• Protein Data Bank: Find experimental IL-6 structures and IL-6 complex structures for template selection and interface inspection.
• MIT OpenCourseWare: Look for free molecular biology, biochemistry, and computational biology course materials that explain protein structure and binding.
• NCBI Resources: Use the NCBI site to explore protein records, conserved domains, and related sequence data.
• UCSF ChimeraX Documentation: Read the free user guides for structure visualization, interface analysis, and surface inspection.