Structure-to-Function: Hypothetical Protein and Binding-Evidence Factory
Automating functional interpretation of poorly characterized pathogen proteins by combining sequence, structure, binding databases, and literature evidence.
Goal
Automate functional interpretation of poorly characterized pathogen proteins by combining sequence, structure, binding databases, and literature evidence.
Three-Day MVP
Analyze 50–100 hypothetical proteins from a selected pathogen group such as Chlamydiales. The agent should retrieve sequences from BV-BRC, run similarity and structure tools, identify structural neighbors, search PDB/BindingDB/ChEMBL evidence, extract supporting statements from papers, and generate ranked functional annotations.
AutoPDB components can create a provenance-aware training dataset of protein–ligand or protein–protein interactions.
Model and Evaluation
Train or calibrate an embedding-based function classifier or evidence-ranking model. Compare recommendations with curated annotations or held-out known proteins.
Team
Team assignments are being finalized ahead of the codeathon. Participants can review project teams, and request reassignment, in the shared participant spreadsheet circulated by the organizing team.