Representing local protein environments with machine learning force fields

conference paper Representing local protein environments with machine learning force fields published yes Meital I Bojan author 11d88cf5-91ca-11f0-a95f-edf9f08f47b7 Sanketh Vedula author Sai A Maddipatla author e957f5e5-91c9-11f0-a95f-e090f66ecb4d Nadav E Sellam author ef280fe0-91c9-11f0-a95f-8dea3f5bc513 Anar Rzayev author 2cd60677-9acd-11f1-ae1a-a85ae1c4dd35 Federico Napoli author d42e08e7-f4fc-11eb-af0a-d71e26138f1b0000-0002-9043-136X Paul Schanda author 7B541462-FAF6-11E9-A490-E8DFE56974250000-0002-9350-7606 Alexander Bronstein author 58f3726e-7cba-11ef-ad8b-e6e8cb3904e60000-0001-9699-8730 GradSch department PaSc department AlBr department ICLR: International Conference on Learning Representations The local structure of a protein strongly impacts its function and interactions with other molecules. Representing local biomolecular environments remains a key challenge while applying machine learning approaches over protein structures. The structural and chemical variability of these environments makes them challenging to model, and performing representation learning on these objects remains largely under-explored. In this work, we propose representations for local protein environments that leverage intermediate features from machine learning force fields (MLFFs). We extensively benchmark state-of-the-art MLFFs, comparing their performance across latent spaces and downstream tasks, and show that their embeddings capture local structural (e.g., secondary motifs) and chemical features (e.g., amino acid identity and protonation state), organizing protein environments into a structured manifold. We show that these representations enable zero-shot generalization and transfer across diverse downstream tasks. As a case study, we build a physics-informed, uncertainty-aware chemical shift predictor that achieves state-of-the-art accuracy in biomolecular NMR spectroscopy. Our results establish MLFFs as general-purpose, reusable representation learners for protein modeling, opening new directions in representation learning for structured physical systems. Code and data are available at https://github.com/mb012/MLFF_representation. https://research-explorer.ista.ac.at/download/22722/22723/2026_ICLR_Bojan.pdf application/pdfno 2026Rio de Janeiro, Brazil eng 14th International Conference on Learning Representations 2505.23354 2026100760-100799 https://github.com/mb012/MLFF_representation M.I. Bojan, S. Vedula, S.A. Maddipatla, N.E. Sellam, A. Rzayev, F. Napoli, P. Schanda, A.M. Bronstein, in:, 14th International Conference on Learning Representations, 2026, pp. 100760–100799. M. I. Bojan et al., “Representing local protein environments with machine learning force fields,” in 14th International Conference on Learning Representations, Rio de Janeiro, Brazil, 2026, vol. 2026, pp. 100760–100799. Bojan MI, Vedula S, Maddipatla SA, Sellam NE, Rzayev A, Napoli F, Schanda P, Bronstein AM. 2026. Representing local protein environments with machine learning force fields. 14th International Conference on Learning Representations. ICLR: International Conference on Learning Representations vol. 2026, 100760–100799. Bojan, Meital I, Sanketh Vedula, Sai A Maddipatla, Nadav E Sellam, Anar Rzayev, Federico Napoli, Paul Schanda, and Alex M. Bronstein. “Representing Local Protein Environments with Machine Learning Force Fields.” In 14th International Conference on Learning Representations, 2026:100760–99, 2026. Bojan MI, Vedula S, Maddipatla SA, et al. Representing local protein environments with machine learning force fields. In: 14th International Conference on Learning Representations. Vol 2026. ; 2026:100760-100799. Bojan, M. I., Vedula, S., Maddipatla, S. A., Sellam, N. E., Rzayev, A., Napoli, F., … Bronstein, A. M. (2026). Representing local protein environments with machine learning force fields. In 14th International Conference on Learning Representations (Vol. 2026, pp. 100760–100799). Rio de Janeiro, Brazil. Bojan, Meital I., et al. “Representing Local Protein Environments with Machine Learning Force Fields.” 14th International Conference on Learning Representations, vol. 2026, 2026, pp. 100760–99. 227222026-08-17T12:03:24Z2026-08-18T06:36:55Z