Graph Neural Network Force Fields (GPTFF-mol) for Organic Molecules from Optimization Trajectories (OpenGEM26)

2026-07-23

AI Summary

This work addresses the high computational cost of conventional density functional theory (DFT) and the limited accuracy and generalization of existing machine learning force fields for sulfur- and chlorine-containing organic molecules and non-equilibrium configurations. To this end, the authors construct OpenGEM26, a large-scale dataset comprising 200,000 unique molecules, 4.4 million conformations, and their full geometry optimization trajectories, systematically incorporating diverse non-equilibrium structures of sulfur- and chlorine-containing systems for the first time, thereby substantially expanding conformational coverage. Trained on ωB97X-D/Def2-SVP(TZVP) DFT data, the proposed graph neural network force field, GPTFF-mol, achieves a mean absolute error of 16 meV per molecule (0.82 meV per atom) in energy prediction and demonstrates superior performance in force prediction and molecular dynamics simulations compared to ANI-2x, with high accuracy and reliability validated on tasks such as butane rotation and keto-enol tautomerization.

Abstract

Density functional theory (DFT) serves as a reliable tool for atomistic molecular simulations, while machine learning potentials have become powerful complements to balance accuracy and efficiency. In this work, we release OpenGEM26 (Open Generated Ensemble of Molecules, 2026), a large-scale dataset comprising 200,000 unique molecules and 4.4 million conformations composed of H, C, N, O, S and Cl with up to ten heavy atoms. All calculations are carried out at the ωB97X-D/Def2-SVP and Def2-TZVP levels with dispersion corrections, and complete structural optimization trajectories and abundant non-equilibrium structures are recorded. Statistical analyses confirm that this dataset covers a broader conformational space than QM9 in terms of energy, bond lengths and bond angles. A graph neural network-based potential GPTFF-mol is trained using the new dataset, achieving an energy mean absolute error of 16 meV/molecule, which is equivalent to 0.82meV/atom, and superior force prediction performance compared with ANI-2x. Validated by butane rotation and keto-enol tautomerization tests, the model accurately describes molecular dynamical behaviors and reaction barriers at distorted geometries. This work provides a high-quality resource and robust ML potential for efficient simulations of sulfur- and chlorine-containing organic molecules.

Problem

• machine learning potentials
• organic molecules
• conformational space
• force fields
• non-equilibrium structures

Innovation

• Graph Neural Network
• Machine Learning Force Field
• Conformational Trajectory
• OpenGEM26 Dataset
• Non-equilibrium Structures