Model a peptide on your laptop. Publish it for the world to see.
No lab required. No PhD required. Just a computer and curiosity.
Don’t have any of these? You can still contribute: literature updates, peer reviews, and target suggestions need only a browser.
Two things needed: an AI coding agent and the cookbook.
1. Pick an AI agent — any of these works:
2. Clone the cookbook:
The cookbook contains ready-made recipes for OpenFold3-MLX (Apple Silicon), AlphaFold2-Multimer (CUDA), and Boltz-1 (CUDA). Give it to your agent — they’ll figure it out. Use 2–3 agents for best results; one agent won’t fit the full context.
The easiest way to start: fork an existing card. You inherit the sequence, target, and recipe — then run it yourself.
Your agent takes the recipe, runs the prediction, and produces: a 3D structure file (PDB/CIF), confidence scores (ipTM, ranking score, pLDDT), a reproducibility recipe for others to verify.
| MacBook M4 | ~50 seconds | for a 14-mer on GDF-8 |
| Google Colab T4 | ~3 minutes | for a 14-mer on GDF-8 |
| Linux 4090 | ~30 seconds | for a 14-mer on GDF-8 |
You can stop here. Your prediction is valuable even without synthesis. Upload it and someone else might synthesize it later.
Upload via your agent (automated) or via the web form.
Your card gets: a permanent ID (pep-XXXXX), a public page with 3D viewer, attribution to you, a recipe others can fork, CC-BY-SA 4.0 license.
Upload your card →Your card is live. Other people see it. Things that might happen: someone reproduces your prediction, someone improves it, someone synthesizes the peptide ($80–300 via CRO), someone runs a binding assay, an agent cross-references IEDB, a researcher cites it.
You don’t have to do any of these. Publishing the prediction is enough. The community takes it from there.