ByteDance

A fully atomic protein co-design model using unified multimodal diffusion to jointly refine atom types and coordinates in a single stage, with support for non-canonical amino acids.

Enhanced 464M-parameter version of Protenix with substantial gains in antibody-antigen complex prediction over v1, plus target-conditioned VHH-Fc generative design with up to 48% hit rates.

All-atom generative foundation model trained on 5M+ biomolecular complexes that designs small molecules, peptides, and nanobodies against a target site from one checkpoint.

Open-source PyTorch reproduction of AlphaFold 3 (Apache 2.0) that matches or exceeds AF3 performance on protein-ligand, protein-protein, and protein-nucleic acid benchmarks.

Generative foundation model for cryo-EM density maps using flow matching, enabling zero-shot denoising, map sharpening, and missing wedge restoration.

A holistic evaluation framework for protein foundation models, assessing 25+ models across 8 tasks using four-dimensional metrics: quality, novelty, diversity, and robustness.