Wenzhi Xiao

Protenix-v2 improves structure prediction and biomolecular design, achieving higher success rates in antibody-antigen and GPCR target tasks, with high developability and diversity in resulting hits.

Protenix-v1 is an open-source biomolecular structure prediction model that outperforms AlphaFold3 with the same training data cutoff, offering improved accuracy and new features for real-world applications.

PXDesign achieves high hit rates for protein binder design by combining generative models and filtering strategies, validated experimentally and made publicly available.

The paper introduces PXMeter, a toolkit for evaluating structure prediction models, and benchmarks several models on a curated dataset from the PDB.

An open AlphaFold3 reproduction, with weights, inference and training code, that matches AF3 on ligand and RNA benchmarks, beats AlphaFold-Multimer 2.3 on protein interfaces and flags possible memorisation.