Protenix is ByteDance Research's implementation of the AlphaFold 3 architecture, with targeted refinements to model confidence and initialization that improve prediction quality on complex structures. In protein-peptide complex benchmarks, Protenix leads on binding pocket accuracy — making it particularly well-suited for teams focused on drug design, where precise pocket geometry matters.

Protenix supports modeling of proteins, small molecule ligands, and multi-component complexes as a unified system.

Key capabilities:

• Protein monomer, multimer, and protein-ligand complex prediction

• Leads benchmarks on binding pocket prediction accuracy

• Models proteins and small molecule ligands as a single complex

• GPU-accelerated inference with MSA support in Benchling