Inverse Problems with Experiment-Guided AlphaFold

This talk explores a framework for building experiment-grounded protein structure generative models that capture conformational ensembles consistent with experimental data. Learn how researchers treat state-of-the-art protein structure predictors like AlphaFold3 as sequence-conditioned structural priors and reframe ensemble modeling as posterior inference of protein structures based on experimental measurements. The presentation demonstrates the framework's ability to incorporate various experimental measurements, revealing previously unmodeled conformational heterogeneity from crystallographic densities and generating high-accuracy NMR ensembles significantly faster than current methods. Discover how these ensembles sometimes outperform both AlphaFold3 and publicly deposited structures in the Protein Data Bank in terms of fitting experimental data, potentially revolutionizing our approach to modeling conformational diversity in proteins.