HEIWA: Hybrid Explicit–Implicit Water Attention Affects Force Field Accuracy in Structure-Based Protein Language Model ESM3
Journal of Chemical Information and Modeling · 09.11.2026 · 66 (21) · 13827-13845
DOI: 10.1021/acs.jcim.6c01844 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Chemical Information and Modeling, online erschienen am 23.10.2026. Verfasst von Ngai Hei Ernest Ho und 3 weiteren Autor:innen. Beteiligte Einrichtung: Department of Chemical Engineering, National Cheng Kung University, Tainan701, Taiwan. Hauptthemen laut MEDLINE: Water, Proteins. Die Arbeit nennt 70 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Ministry of Science and Technology, Taiwan.
Abstract (Originalsprache)
Protein language models have achieved remarkable success in structure prediction and de novo design. However, their understanding of physical interactions remains limited. For structure-based models, we hypothesize that this limitation stems from the absence of explicit solvent modeling: water mediates hydrogen bonding, hydrophobic burial, and electrostatic interactions that govern protein thermodynamics and kinetics, yet current structure-based models treat proteins as if they exist in a vacuum. Using thermophilic protein classification as a proxy task for force-field awareness, per-residue logit mis-correlated with H-bonds as a destabilizing attribute and exhibits significant solvent-accessible surface area (SASA)-dependent error, despite high overall classification accuracy. These artifacts point to the water–protein interface as a source of systematic bias. To address this, we propose HEIWA (HybridExplicit-Implicit Water Attention), a novel attention module that injects solvent-mediated force-field information into static ESM3 residue embeddings via an all-query-to-all-key water–residue attention. Biophysics-guided model selection identifies configurations whose residue-level outputs are more consistent with the RMSF benchmark and selected biophysical attributes, while classification performance remains broadly comparable to the baseline. By forcing the model to predict the protein thermophilicity from the contributions of individual residues, HEIWA learns biophysical patterns that can be extended to SUPFAM absent from the training data set. HEIWA also largely preserves the embedding structure of ESM3. Molecular dynamics simulations show improved kinetic stability under thermophilic conditions, suggesting the potential to generate structures with greater kinetic stability. To our knowledge, this is the first study to incorporate explicit solvent encoding into a residue-resolution protein language model.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Substanzen
Autor:innen und Institutionen
- Department of Chemical Engineering, National Cheng Kung University, Tainan701, Taiwan.
- Department of Computer Science, Tunghai University, Taichung407, Taiwan.
- Department of Medical Research, Kuang Tien General Hospital, Taichung433, Taiwan.
- Research Center for Smart Sustainable Circular Economy, Tunghai University, Taichung407, Taiwan.
- Department of Chemical and Material Engineering, Tunghai University, Taichung407, Taiwan.
- Department of Chemical Engineering and Materials Science, Yuan Ze University, Chung-Li302, Taiwan.
ORCID-Profile: Ngai Hei Ernest Ho, Chao Tung Yang, I-Son Ng, Jo-Shu Chang
Förderung
- Ministry of Science and Technology, Taiwan
Zitieren
APA
Ernest Ho, N. H., Yang, C. T., Ng, I. S., & Chang, J. S. (2026). HEIWA: Hybrid Explicit–Implicit Water Attention Affects Force Field Accuracy in Structure-Based Protein Language Model ESM3. Journal of Chemical Information and Modeling, 66(21), 13827-13845. https://doi.org/10.1021/acs.jcim.6c01844
Vancouver
Ernest Ho NH, Yang CT, Ng IS, Chang JS. HEIWA: Hybrid Explicit–Implicit Water Attention Affects Force Field Accuracy in Structure-Based Protein Language Model ESM3. Journal of Chemical Information and Modeling. 2026;66(21):13827-13845. doi:10.1021/acs.jcim.6c01844
BibTeX
@article{ErnestHo2026,
author = {Ernest Ho, Ngai Hei and Yang, Chao Tung and Ng, I-Son and Chang, Jo-Shu},
title = {HEIWA: Hybrid Explicit–Implicit Water Attention Affects Force Field Accuracy in Structure-Based Protein Language Model ESM3},
journal = {Journal of Chemical Information and Modeling},
year = {2026},
volume = {66},
number = {21},
pages = {13827--13845},
doi = {10.1021/acs.jcim.6c01844}
}
EUVIMED