Jun 2025· Journal of Chemical Information and Modeling· Vol 65, pp. 10364-10374· 1 citation· 32 references
MedicineComputer Science
TL;DR
This work introduces an advanced equivariant graph attention neural network specifically engineered to model long-range atomic electrostatic interactions with high precision, and improves the model's accuracy, generalization, and robustness in complex scenarios.
Abstract
Atomic charge is a fundamental quantum chemical property essential for advancing drug design and discovery. Although quantum mechanics (QM) methods offer the highest level of accuracy, their computational demands scale quadratically with the number of atoms, limiting their practicality for large-scale applications. In light of this, empirical and semiempirical methods have been introduced to improve computational efficiency, albeit often at the expense of accuracy. The advent of artificial intelligence has witnessed a growing application of machine learning (ML) techniques to accelerate atomic charge predictions. However, existing ML models often suffer from low accuracy and limited generalization capabilities. To address these challenges, we introduce an advanced equivariant graph attention neural network specifically engineered to model long-range atomic electrostatic interactions with high precision. This model introduces a sophisticated global graph attention mechanism, enabling it to capture charge contributions across multiple scales. By utilizing a combination of structural symmetry-preserving transformations and multiscale attention, our approach not only preserves the inherent symmetries of molecular structures but also substantially improves the model's accuracy, generalization, and robustness in complex scenarios. Our empirical analyses demonstrate that, compared to leading baseline models, the proposed model improves charge prediction accuracy by over 40% on average across various charge-calculation schemes. Remarkably, the model achieves superior performance on the external RESP (restrained electrostatic potential) test data sets, with a 54.6% improvement over the baseline. Additionally, we evaluated our charge model under the setting of virtual screening, where it outperforms both the OPLS3 charges and baseline deep learning models across all evaluation metrics, highlighting its extensive potential for scientific discovery.
DeGAT enables efficient and accurate prediction of partial atomic charges in MOFs while maintaining charge neutrality and physical consistency, providing a scalable parametrization scheme for high-throughput screening and molecular simulations of porous materials.
Yanhui Sun, Ze-Heng Yu, Yu-Hua Dong et al.· Journal of Chemical Theory a...· 0 citations
Machine learning potentials (MLPs) have emerged as game-changing tools for large-scale atomic simulations, overcoming the poor-scaling limitation intrinsic to traditional quantum mechanics (QM) methods. However, accurately incorporating electronic information remains a significant challenge for MLPs, particularly in efficiently computing the dynamic properties of matter under electric fields─a task central to topics such as infrared spectroscopy, interfaces under electric fields, and ferroelectric polarization. Herein, we report a physics-informed pairwise charge-transfer (PQT) theory to derive dynamic equations for macroscopic polarization that inherently conserve fundamental physical laws. Using the PQT theory, a Generalized Global Neural Network (GGNN) enhanced with the PQT mechanism is developed for the rapid prediction of dynamic properties under electric fields, applicable to both molecules and materials across the periodic table. Specifically, a generalized global data set comprising 3.18 million structures with atomic charges for 81 elements is utilized to pretrain a GGNN patched with PQT modules. Leveraging this pretrained GGNN-PQT potential, we can conveniently sample the potential energy surface under electric fields and fine-tune the potential using a small QM data set containing exact response properties at low cost. Our GGNN-PQT has linear scaling and introduces a low computational overhead compared to the standard GGNN, yet achieves both high speeds and low scaling. We demonstrate the performance of GGNN-PQT in computing dynamic response properties across a wide range of systems, including isolated molecules, adsorbed molecules, molecular crystals, liquid water, and ferroelectric materials.
Xin-Tian Xie, Zhen-Xiong Wang, Zhen-Xing Yang et al.· Journal of Chemical Theory a...· 0 citations
Conventional local machine-learning interatomic potentials describe atomic environments with a finite cutoff radius and therefore have difficulty capturing long-range electrostatic coupling in polar, charged, or charge-transfer systems. This paper proposes CE-ETNet (Charge-Equilibration-Enhanced Equivariant Transformer Network), an equivariant Transformer interatomic potential constrained by charge equilibration. The model learns local chemical environments formed by atom types, geometric edges, and radial basis features through an equivariant representation encoder, and predicts atomic electronegativities. Under the constraint of total charge conservation, a differentiable charge equilibration procedure is used to solve partial charges and electrostatic energy, thereby incorporating long-range Coulomb interactions into energy and force prediction. On three benchmark datasets, QM9, revised MD17, and tmQM_wB97MV, CE-ETNet reduces the mean absolute error (MAE) of energy and force prediction by about 8% on average compared with existing models, reaching chemical accuracy. The experimental results show that explicit long-range electrostatic modeling improves the predictive performance of equivariant graph neural network interatomic potentials and provides a useful design direction for machine-learning interatomic potential models.
Yifei Shi, Tao Luo· Computers and artificial int...· 0 citations
GeoNet is a physicochemical-principle-guided framework for modeling dual-range atomic interactions that achieves the smallest model size and the shortest training time, demonstrating both superior predictive performance and computational efficiency.
Mandala is a modular software framework for learning block-sparse electronic-structure matrices with E(3)-equivariant graph neural networks that connects electronic-structure learning and observable-guided modeling while retaining a representation tied to quantum-mechanical operators rather than only scalar or vector targets as in MLIPs.
B. Brzoza, Wiktoria Szopa, Z. Elabid et al.· 0 citations
This work proposes an operator-centric framework in which the external (nuclear) potential, expressed in an AO basis, serves as the model input and builds hierarchical, body-ordered representations of atomic configurations that closely mirror the principles underlying several popular atom-centered descriptors.
Jigyasa Nigam, T. Smidt, G. Dusson· Journal of Chemical Physics· 2 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduAug 18, 2026
A new method for surgically removing training examples from a model reveals that as datasets grow, the link between what a model learns and what it produces dissolves.
What does it take to trust AI-driven HVAC optimization? Our AI Model Factory combines agents, machine learning, reinforcement learning and deterministic checks in a governed workflow designed for messy, real-world building data. The post We built an AI factory for HVAC control appeared first on GPT-Lab.
