Representing local protein environments with machine learning force fields

TL;DR

This paper introduces a novel machine learning-based representation of local protein environments derived from atomistic foundation models, capturing structural and chemical features effectively, and enabling improved biomolecular modeling and NMR chemical shift prediction.

Contribution

The work presents a new AFM-derived embedding for local protein environments that captures structural and chemical features, and enables a physics-informed chemical shift predictor with state-of-the-art accuracy.

Findings

Effective embedding captures local structure and chemical features.

Representation space exhibits meaningful structure for biomolecular environments.

Enables a state-of-the-art chemical shift prediction in NMR.

Abstract

The local structure of a protein strongly impacts its function and interactions with other molecules. Therefore, a concise, informative representation of a local protein environment is essential for modeling and designing proteins and biomolecular interactions. However, these environments' extensive structural and chemical variability makes them challenging to model, and such representations remain under-explored. In this work, we propose a novel representation for a local protein environment derived from the intermediate features of atomistic foundation models (AFMs). We demonstrate that this embedding effectively captures both local structure (e.g., secondary motifs), and chemical features (e.g., amino-acid identity and protonation state). We further show that the AFM-derived representation space exhibits meaningful structure, enabling the construction of data-driven priors over the distribution of biomolecular environments. Finally, in the context of biomolecular NMR spectroscopy, we demonstrate that the proposed representations enable a first-of-its-kind physics-informed chemical shift predictor that achieves state-of-the-art accuracy. Our results demonstrate the surprising effectiveness of atomistic foundation models and their emergent representations for protein modeling beyond traditional molecular simulations. We believe this will open new lines of work in constructing effective functional representations for protein environments.