chai-lab

What problem does it solve?

Enables end-to-end prediction of biomolecular structures for complexes and assemblies using multi-modal models, reducing manual trial-and-error in structure generation.

Core Features & Use Cases

• Multi-modal all-atom structure prediction for protein–protein, protein–ligand, protein–DNA/RNA, glycosylated proteins, and binder validation scenarios.
• Supports optional MSAs, templates, and restraints to guide interface geometry and confidence scoring.
• Provides installation, CLI (fold, fold-batch, a3m-to-pqt, citation), and Python API (run_inference, run_folding_on_context) workflows, plus FASTA-like input formats and restraint specifications.
• Outputs include CIF structures and per-sample NPZ scores for downstream ranking and QC, suitable for batch campaigns.

Quick Start

Provide a FASTA-like input with all chains and run chai-lab fold to produce CIFs, scores, and MSAs outputs in an output directory.