FlowBack-Adjoint : Physics-Aware and Energy-Guided Conditional Flow-Matching for Protein Backmapping
Journal of Chemical Information and Modeling · 17.11.2026
DOI: 10.1021/acs.jcim.6c02114 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Chemical Information and Modeling, online erschienen am 17.11.2026. Verfasst von Alex Berlaga und 4 weiteren Autor:innen. Die Arbeit nennt 87 Literaturangaben. Gefördert von: National Science Foundation, U.S. Department of Health and Human Services.
Abstract (Originalsprache)
Coarse-grained (CG) molecular models of proteins can substantially increase the time and length scales accessible to molecular dynamics simulations of proteins, but recovery of accurate all-atom (AA) ensembles from CG simulation trajectories can be essential for exposing molecular mechanisms and for the calculation of physical properties requiring atomistic detail. The recently reported deep generative model FlowBack restores AA detail to protein Cα traces using a flow-matching architecture and demonstrates state-of-the-art performance in backmapping of AA structural ensembles (J. Chem. Inf. Model. 65, 2, 672–692, 2025). Training, however, is performed exclusively on structural data, and in the absence of any awareness of interatomic energies or forces, small structural imperfections can produce high-energy backmapped structures. In this work, we introduce FlowBack-Adjoint as a lightweight enhancement that upgrades the pretrained FlowBack model through a one-time, physics-aware post-training pass. Auxiliary contributions to the flow introduce physical awareness of bond lengths, bond angles, and Lennard-Jones interactions, and gradients of a molecular mechanics force field energy are incorporated via adjoint matching to steer FlowBack-Adjoint toward low-energy configurations without compromising structural fidelity or diversity. In benchmark tests against FlowBack, FlowBack-Adjoint lowers single-point energies by a median of ∼30 kcal/mol·residue, reduces errors in bond lengths by >95%, eliminates >97% of molecular clashes, maintains excellent diversity of the AA configurational ensemble, and produces configurations capable of initializing stable all-atom molecular dynamics simulations. In applications to large protein–protein interfaces, including peptide–major histocompatibility complexes (peptide–MHCs) and HIV capsid assemblies, the model improves agreement with molecular dynamics-derived contact frequencies and increases the recall of experimentally observed interactions. We propose FlowBack-Adjoint as an accurate and efficient physics-aware deep generative model for AA backmapping from Cα traces generated by structure prediction tools, coarse-grained simulations, or low-resolution experimental data.
Förderung
- National Science Foundation
- U.S. Department of Health and Human Services
Zitieren
APA
Berlaga, A., Jones, M. S., Waltmann, C., Voth, G. A., & Ferguson, A. L. (2026). FlowBack-Adjoint : Physics-Aware and Energy-Guided Conditional Flow-Matching for Protein Backmapping. Journal of Chemical Information and Modeling. https://doi.org/10.1021/acs.jcim.6c02114
Vancouver
Berlaga A, Jones MS, Waltmann C, Voth GA, Ferguson AL. FlowBack-Adjoint : Physics-Aware and Energy-Guided Conditional Flow-Matching for Protein Backmapping. Journal of Chemical Information and Modeling. 2026. doi:10.1021/acs.jcim.6c02114
BibTeX
@article{Berlaga2026,
author = {Berlaga, Alex and Jones, Michael S. and Waltmann, Curt and Voth, Gregory A. and Ferguson, Andrew L.},
title = {FlowBack-Adjoint : Physics-Aware and Energy-Guided Conditional Flow-Matching for Protein Backmapping},
journal = {Journal of Chemical Information and Modeling},
year = {2026},
doi = {10.1021/acs.jcim.6c02114}
}
EUVIMED