{"publication_status":"published","intvolume":"      2026","date_published":"2026-05-01T00:00:00Z","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","researchdata_availability":"yes","arxiv":1,"year":"2026","publication":"14th International Conference on Learning Representations","das_tickbox":"1","oa_version":"Published Version","dataavailabilitystatement":"The code, trained models, and data-processing scripts are publicly available at https://github.\r\ncom/mb012/MLFF_representation. In addition, complete details of the models and optimization parameters are provided in Appendix G.3. The hardware resources used to produce the\r\nresults are specified in Appendix I.4. The loss functions, evaluation metrics, and details regarding\r\nablation studies are specified in Appendix G. These details ensure that all results reported in the paper\r\ncan be independently verified.","abstract":[{"lang":"eng","text":"The local structure of a protein strongly impacts its function and interactions\r\nwith other molecules. Representing local biomolecular environments remains a\r\nkey challenge while applying machine learning approaches over protein structures. The structural and chemical variability of these environments makes them\r\nchallenging to model, and performing representation learning on these objects\r\nremains largely under-explored. In this work, we propose representations for\r\nlocal protein environments that leverage intermediate features from machine learning force fields (MLFFs). We extensively benchmark state-of-the-art MLFFs,\r\ncomparing their performance across latent spaces and downstream tasks, and\r\nshow that their embeddings capture local structural (e.g., secondary motifs) and\r\nchemical features (e.g., amino acid identity and protonation state), organizing\r\nprotein environments into a structured manifold. We show that these representations enable zero-shot generalization and transfer across diverse downstream\r\ntasks. As a case study, we build a physics-informed, uncertainty-aware chemical shift predictor that achieves state-of-the-art accuracy in biomolecular NMR\r\nspectroscopy. 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\r\nhttps://github.com/mb012/MLFF_representation.\r\n"}],"ddc":["000","570"],"file_date_updated":"2026-08-18T06:33:14Z","status":"public","oa":1,"date_updated":"2026-08-18T06:36:55Z","author":[{"id":"11d88cf5-91ca-11f0-a95f-edf9f08f47b7","first_name":"Meital I","full_name":"Bojan, Meital I","last_name":"Bojan"},{"first_name":"Sanketh","last_name":"Vedula","full_name":"Vedula, Sanketh"},{"last_name":"Maddipatla","full_name":"Maddipatla, Sai A","id":"e957f5e5-91c9-11f0-a95f-e090f66ecb4d","first_name":"Sai A"},{"last_name":"Sellam","full_name":"Sellam, Nadav E","first_name":"Nadav E","id":"ef280fe0-91c9-11f0-a95f-8dea3f5bc513"},{"last_name":"Rzayev","full_name":"Rzayev, Anar","first_name":"Anar","id":"2cd60677-9acd-11f1-ae1a-a85ae1c4dd35"},{"last_name":"Napoli","full_name":"Napoli, Federico","orcid":"0000-0002-9043-136X","id":"d42e08e7-f4fc-11eb-af0a-d71e26138f1b","first_name":"Federico"},{"first_name":"Paul","id":"7B541462-FAF6-11E9-A490-E8DFE5697425","orcid":"0000-0002-9350-7606","last_name":"Schanda","full_name":"Schanda, Paul"},{"orcid":"0000-0001-9699-8730","first_name":"Alexander","id":"58f3726e-7cba-11ef-ad8b-e6e8cb3904e6","full_name":"Bronstein, Alexander","last_name":"Bronstein"}],"month":"05","title":"Representing local protein environments with machine learning force fields","date_created":"2026-08-17T12:03:24Z","citation":{"mla":"Bojan, Meital I., et al. “Representing Local Protein Environments with Machine Learning Force Fields.” <i>14th International Conference on Learning Representations</i>, vol. 2026, 2026, pp. 100760–99.","chicago":"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 <i>14th International Conference on Learning Representations</i>, 2026:100760–99, 2026.","ama":"Bojan MI, Vedula S, Maddipatla SA, et al. Representing local protein environments with machine learning force fields. In: <i>14th International Conference on Learning Representations</i>. Vol 2026. ; 2026:100760-100799.","apa":"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 <i>14th International Conference on Learning Representations</i> (Vol. 2026, pp. 100760–100799). Rio de Janeiro, Brazil.","ieee":"M. I. Bojan <i>et al.</i>, “Representing local protein environments with machine learning force fields,” in <i>14th International Conference on Learning Representations</i>, Rio de Janeiro, Brazil, 2026, vol. 2026, pp. 100760–100799.","short":"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.","ista":"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."},"tmp":{"image":"/images/cc_by.png","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode"},"acknowledged_ssus":[{"_id":"ScienComp"}],"corr_author":"1","volume":2026,"OA_place":"publisher","page":"100760-100799","type":"conference","article_processing_charge":"No","department":[{"_id":"GradSch"},{"_id":"PaSc"},{"_id":"AlBr"}],"related_material":{"link":[{"relation":"software","url":"https://github.com/mb012/MLFF_representation"}]},"OA_type":"gold","acknowledgement":"This work was supported by the Institute of Science and Technology Austria (ISTA) through the IPC\r\ngrant “Generative Protein NMR” and by the Israeli Science Foundation (ISF) under grant number\r\n1834/24. This research used resources of the Institute of Science and Technology Austria’s scientific\r\ncomputing cluster. S.V. was supported in part by funding from the Eric and Wendy Schmidt Center at\r\nthe Broad Institute of MIT and Harvard.","supplementarymaterial":"yes","_id":"22722","has_accepted_license":"1","day":"01","language":[{"iso":"eng"}],"quality_controlled":"1","file":[{"file_id":"22723","file_size":8534339,"content_type":"application/pdf","date_created":"2026-08-18T06:33:14Z","success":1,"checksum":"9f43f5469443388ec55388d95c4cf243","access_level":"open_access","file_name":"2026_ICLR_Bojan.pdf","creator":"dernst","relation":"main_file","date_updated":"2026-08-18T06:33:14Z"}],"conference":{"end_date":"2026-04-27","location":"Rio de Janeiro, Brazil","start_date":"2026-04-23","name":"ICLR: International Conference on Learning Representations"},"external_id":{"arxiv":["2505.23354"]}}