pdb2reaction : End-to-End Reaction-Path Elucidation from PDB Structures Using Machine-Learning Interatomic Potentials
- 1The University of Tokyo , , 1-1-1 Yayoi , Bunkyo-Ku, ,
Abstract
Elucidating enzymatic reaction mechanisms requires a sequence of computational tasks comprising active-site extraction, minimum-energy-path search, transition-state (TS) refinement, intrinsic reaction coordinate (IRC) validation, and quasi-rigid-rotor harmonic-oscillator (QRRHO) thermochemistry─stages typically connected by ad hoc scripting and per-system tuning. We present pdb2reaction, an open-source Python command-line toolkit that automates this entire pipeline directly from a user-curated PDB using a single machine-learning interatomic potential (MLIP) backend. A GPU-accelerated pysisyphus fork is bundled to perform the Hessian-based heavy computations─Hessian, IRC, and vibrational analysis─on the same CUDA device as the MLIP for the default backend, minimizing data transfer overhead. This toolkit also implements a bond-change-driven recursive path-search algorithm, which recovered the two-step reaction mechanism of the geranyl pyrophosphate C6-methyltransferase BezA─an SN2-like methyl transfer followed by glutamate-mediated deprotonation via a cationic intermediate─locating both transition states and the intermediate itself without a user-supplied intermediate geometry: from a reactant-state active-site cluster and a minimal scan list, and independently from the reactant and product structures alone. On a benchmark of 23 reaction steps across six enzymes, the broadest-coverage backend, UMA-m-1.1, reproduced the literature mechanism for 19 steps, each with a single imaginary frequency at the transition state. For BezA, all five backends recovered the same two-step pathway and the same rate-limiting step. pdb2reaction enables rapid PDB-to-mechanism elucidation on a single GPU, providing a high-throughput complement to higher-cost DFT-based cluster-model calculations and full enzyme–solvent QM/MM calculations.
Versions
Related
- Benchmarking machine-learning interatomic potentials using classical force fields
- Why Multi-Layer Message Passing Works: Completeness Theory for Graph Neural Network Interatomic Potentials
- MLIP Detective: Active Failure Mode Discovery Beyond Benchmark Scores for Machine-Learning Interatomic Potentials
- Cluster-based Structural Similarity for Dataset Visualization and Data Selection for Machine Learning Interatomic Potentials
- BranchIP: Learning Adaptive Equivariant Computation for Interatomic Potentials
BibTeX
@article{ohmura2026pdb2reaction,
title = {{pdb2reaction : End-to-End Reaction-Path Elucidation from PDB Structures Using Machine-Learning Interatomic Potentials}},
author = {Takuto Ohmura and Hajime Sato and Tohru Terada},
journal = {ACS Omega},
year = {2026},
month = oct,
doi = {10.1021/acsomega.6c08242},
url = {https://pubs.acs.org/acsodf/article/11/42/65454/5446631/pdb2reaction-End-to-End-Reaction-Path-Elucidation},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - pdb2reaction : End-to-End Reaction-Path Elucidation from PDB Structures Using Machine-Learning Interatomic Potentials
AU - Takuto Ohmura
AU - Hajime Sato
AU - Tohru Terada
PY - 2026
DA - 2026/10/07
JO - ACS Omega
DO - 10.1021/acsomega.6c08242
UR - https://pubs.acs.org/acsodf/article/11/42/65454/5446631/pdb2reaction-End-to-End-Reaction-Path-Elucidation
AB - Elucidating enzymatic reaction mechanisms requires a sequence of computational tasks comprising active-site extraction, minimum-energy-path search, transition-state (TS) refinement, intrinsic reaction coordinate (IRC) validation, and quasi-rigid-rotor harmonic-oscillator (QRRHO) thermochemistry─stages typically connected by ad hoc scripting and per-system tuning. We present pdb2reaction, an open-source Python command-line toolkit that automates this entire pipeline directly from a user-curated PDB using a single machine-learning interatomic potential (MLIP) backend. A GPU-accelerated pysisyphus fork is bundled to perform the Hessian-based heavy computations─Hessian, IRC, and vibrational analysis─on the same CUDA device as the MLIP for the default backend, minimizing data transfer overhead. This toolkit also implements a bond-change-driven recursive path-search algorithm, which recovered the two-step reaction mechanism of the geranyl pyrophosphate C6-methyltransferase BezA─an SN2-like methyl transfer followed by glutamate-mediated deprotonation via a cationic intermediate─locating both transition states and the intermediate itself without a user-supplied intermediate geometry: from a reactant-state active-site cluster and a minimal scan list, and independently from the reactant and product structures alone. On a benchmark of 23 reaction steps across six enzymes, the broadest-coverage backend, UMA-m-1.1, reproduced the literature mechanism for 19 steps, each with a single imaginary frequency at the transition state. For BezA, all five backends recovered the same two-step pathway and the same rate-limiting step. pdb2reaction enables rapid PDB-to-mechanism elucidation on a single GPU, providing a high-throughput complement to higher-cost DFT-based cluster-model calculations and full enzyme–solvent QM/MM calculations.
SN - 2470-1343
ER -