Efficient and Accurate Ligand Strain Calculations in Solution with the AIMNet2 Neural Network Potential
Journal of Chemical Information and Modeling · 09.11.2026 · 66 (21) · 14012-14024
DOI: 10.1021/acs.jcim.6c02468 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Chemical Information and Modeling, online erschienen am 21.10.2026. Verfasst von Hatice Gokcan und 1 weiteren Autor:innen. Beteiligte Einrichtung: Department of Chemistry, Mellon College of Science, Carnegie Mellon University, Pittsburgh, Pennsylvania15213, United States. Hauptthemen laut MEDLINE: Neural Networks, Computer. Die Arbeit nennt 84 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Division of Chemistry.
Abstract (Originalsprache)
Conformational strain, the energy a ligand pays to adopt its bound conformation, contributes to binding affinity but is routinely misestimated. The standard estimate references a single gas-phase energy minimum. This ignores solvent and mishandles charged ligands: in vacuo their intramolecular electrostatics collapse them into compact minima that deflate the unbound reference and inflate the apparent strain. We introduce AIMNet2-CPCM, a neural network potential trained to B97-3c energies and forces in a conductor-like continuum solvent (CPCM). It treats neutral and charged species on the same footing, and its batched GPU inference makes a solvated global-minimum search practical across an entire screening library. Across ∼8K PDB ligands the solvated strain has a median of 4.0 kcal/mol and rises with net charge. The trend is well determined through a net charge of −2. The −3 and −4 polyanions dominate the high-strain tail, where continuum solvation is least reliable, and their values are reported as upper bounds. Gas-phase force fields reproduce the neutral strains but overstate the charged ones, because they collapse the unbound minimum that solvent screening keeps open. Adding a continuum solvent to the force field removes most of this error and confirms the mechanism. Profiling at this scale further shows that the same ligand strains very differently across the targets that bind it: ligand strain is a property of the bound pose, not a transferable molecular descriptor.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Substanzen
Autor:innen und Institutionen
- Department of Chemistry, Mellon College of Science, Carnegie Mellon University, Pittsburgh, Pennsylvania15213, United States.
- Department of Materials Science and Engineering, Carnegie Mellon University, Pittsburgh, Pennsylvania15213, United States.
ORCID-Profile: Olexandr Isayev
Förderung
- Division of Chemistry
Zitieren
APA
Gokcan, H., & Isayev, O. (2026). Efficient and Accurate Ligand Strain Calculations in Solution with the AIMNet2 Neural Network Potential. Journal of Chemical Information and Modeling, 66(21), 14012-14024. https://doi.org/10.1021/acs.jcim.6c02468
Vancouver
Gokcan H, Isayev O. Efficient and Accurate Ligand Strain Calculations in Solution with the AIMNet2 Neural Network Potential. Journal of Chemical Information and Modeling. 2026;66(21):14012-14024. doi:10.1021/acs.jcim.6c02468
BibTeX
@article{Gokcan2026,
author = {Gokcan, Hatice and Isayev, Olexandr},
title = {Efficient and Accurate Ligand Strain Calculations in Solution with the AIMNet2 Neural Network Potential},
journal = {Journal of Chemical Information and Modeling},
year = {2026},
volume = {66},
number = {21},
pages = {14012--14024},
doi = {10.1021/acs.jcim.6c02468}
}
EUVIMED