Generalized Multi-State Protein Design with AlphaFold3

Xiwei Cheng, Pengkang Guo, Seonghwan Seo, Casper A. Goverde, Wengong Jin

Keywords: Multi-State Protein Design, Protein Ensemble Generation

Abstract: While recent advances in deep learning have revolutionized single-state protein design, engineering proteins that can adopt multiple distinct structural states remains a fundamental challenge. In this work, we present a generalized framework for multi-state protein design leveraging the advanced all-atom co-folding capabilities of AlphaFold3. We first introduce AF3-Cluster, an evaluation framework that synergizes Multiple Sequence Alignment (MSA) clustering with AlphaFold3 to resolve complex structural ensembles across diverse biochemical contexts. Validation on a benchmark of ten multi-state proteins demonstrates that AF3-Cluster robustly recovers alternative conformations, particularly for binder-mediated transitions where previous single-chain methods often fail. Building on this, we propose AF3-MSD, an evolutionary framework that utilizes AlphaFold3 as a computationally efficient scoring proxy to guide the sculpting of multi-basin energy landscapes. AF3-MSD successfully drives candidates toward higher scores to design functional sequences for seven multi-state proteins, achieving structural agreement on par with natural multi-conformational proteins.

Submission Number: 93