Representing local protein environments with atomistic foundation models

Researchers developed a novel method to represent local protein environments using atomistic foundation models (AFMs), creating embeddings that capture both structural and chemical features. The approach enables construction of data-driven priors for biomolecular environments and achieves state-of-the-art accuracy in physics-informed chemical shift prediction for NMR spectroscopy.

• →Atomistic foundation models can generate effective representations of local protein environments beyond traditional molecular simulations.
• →The AFM-derived embeddings capture both local structure like secondary motifs and chemical features including amino-acid identity.
• →The representation space exhibits meaningful structure that enables construction of data-driven priors over biomolecular environments.
• →A first-of-its-kind physics-informed chemical shift predictor achieved state-of-the-art accuracy using these representations.
• →This work opens new research directions for creating functional representations of protein environments.