Leveraging state-of-the-art AI structure predictors to generate atomic resolution ensembles of intrinsically disordered proteins

Intrinsically disordered proteins (IDPs) play crucial roles in various biological processes. However, their function relies on dynamic conformational ensembles rather than a single structure. Although artificial intelligence has significantly advanced protein structure prediction, generating these conformational ensembles with atomic resolution remains a hurdle.

Recently, researchers have developed bAIes, a framework that utilizes the structural information encoded in AlphaFold2 distograms to produce conformational ensembles of IDPs at atomic resolution. This study expands bAIes' application to six leading structure prediction tools, including AlphaFold2, AlphaFold3, Boltz-1, Boltz-2, ESMFold, and ESMFold2, and systematically compares the structural information encoded in their inter-residue distance distributions.

Despite the disparities between predictors, the study reveals that bAIes generates ensembles with comparable accuracy to experimental NMR and SAXS data, and maintains competitive performance against state-of-the-art all-atom simulations and alternative IDP ensemble generation methods. These findings indicate that bAIes is resilient to the selection of structure prediction tools, and that the distance distributions generated by various deep learning models hold valuable insights into the conformational preferences of disordered proteins.

In a broader context, this research underscores the potential of merging AI-derived structural information with physics-based modeling to create atomistic ensembles of intrinsically disordered proteins and regions.