protenix

What problem does it solve?

Predict high-accuracy 3D structures for protein, nucleic acid, ligand, and ion complexes to validate designs and inform experimental prioritization, while providing confidence metrics for ranking candidates.

Core Features & Use Cases

• AF3 reproduction: ByteDance's Protenix provides AlphaFold 3-level accuracy in an open-source PyTorch implementation for structure prediction.
• Flexible input and entity support: Accepts FASTA and Protenix JSON inputs and handles protein, DNA, RNA, ligand (SMILES), and ion entities.
• MSA-free fast mode and multi-seed ensembles: Run rapid predictions without MSA for high-throughput campaigns or enable MSA for higher accuracy and use multi-seed ensembles for robust confidence estimates.
• Outputs and integration: Produces CIF structure files and confidence JSON (pLDDT, pTM, ipTM) and integrates with downstream QC and campaign skills for filtering and ranking.

Quick Start

Run Protenix on my complex.faa with seeds 42,43,44 using no-use-msa to generate CIF models and a confidence JSON for ranking.