Abstract:Geometric deep learning has been revolutionizing the molecular modeling field. Despite the state-of-the-art neural network models are approaching ab initio accuracy for molecular property prediction, their applications, such as drug discovery and molecular dynamics (MD) simulation, have been hindered by insufficient utilization of geometric information and high computational costs. Here we propose an equivariant geometry-enhanced graph neural network called ViSNet, which elegantly extracts geometric features and efficiently models molecular structures with low computational costs. Our proposed ViSNet outperforms state-of-the-art approaches on multiple MD benchmarks, including MD17, revised MD17 and MD22, and achieves excellent chemical property prediction on QM9 and Molecule3D datasets. Additionally, ViSNet achieved the top winners of PCQM4Mv2 track in the OGB-LCS@NeurIPS2022 competition. Furthermore, through a series of simulations and case studies, ViSNet can efficiently explore the conformational space and provide reasonable interpretability to map geometric representations to molecular structures.

What problem does this paper attempt to address?

The problems that this paper attempts to solve are the insufficient utilization of geometric information and the excessively high computational cost in molecular modeling. Specifically: 1. **Full utilization of geometric information**: Although existing neural network models have approached the accuracy of ab initio in molecular property prediction, in applications such as drug discovery and molecular dynamics (MD) simulations, the performance is limited due to the failure to fully extract and utilize geometric information. 2. **High computational cost**: Traditional molecular modeling methods such as density functional theory (DFT) can accurately calculate the energy, force, and other chemical properties of molecules, but their computational requirements are large. Especially when dealing with large - scale molecular systems or high - precision calculations, a large amount of computational resources and time are required. To solve these problems, the author proposes an equivariant geometry - enhanced graph neural network named ViSNet. ViSNet solves the above problems in the following ways: - **Efficient extraction of geometric features**: ViSNet introduces the Runtime Geometric Computation (RGC) strategy to implicitly extract various geometric features (such as angles, dihedral angles, and improper angles), and has a linear - time complexity, thereby significantly accelerating model training and inference while reducing memory consumption. - **Vector - Scalar Interaction Message - Passing mechanism (ViS - MP)**: ViSNet designs an effective vector - scalar interaction message - passing mechanism to fully utilize geometric information and enhance the interaction between scalars and vectors, so as to better learn the geometric representation of molecular structures. Through these innovations, ViSNet not only achieves state - of - the - art performance on multiple benchmark datasets but also demonstrates high efficiency and high fidelity in molecular dynamics simulations. In addition, ViSNet also shows reasonable interpretability and can map geometric representations onto molecular structures. In summary, this paper aims to improve the utilization rate of geometric information in molecular modeling and reduce the computational cost by proposing ViSNet, thereby promoting the development of fields such as drug discovery and molecular dynamics simulations.

ViSNet: an equivariant geometry-enhanced graph neural network with vector-scalar interactive message passing for molecules

Enhancing geometric representations for molecules with equivariant vector-scalar interactive message passing

ViSNet: a scalable and accurate geometric deep learning potential for molecular dynamics simulation

An Image-enhanced Molecular Graph Representation Learning Framework

An ensemble of VisNet, Transformer-M, and pretraining models for molecular property prediction in OGB Large-Scale Challenge @ NeurIPS 2022

Long-Short-Range Message-Passing: A Physics-Informed Framework to Capture Non-Local Interaction for Scalable Molecular Dynamics Simulation

Long-Short-Range Message-Passing: A Fragmentation-Based Framework to Capture Non-Local Atomistic Interactions

A Universal Framework for Accurate and Efficient Geometric Deep Learning of Molecular Systems

Molecule Graph Networks with Many-body Equivariant Interactions

Heterogeneous Relational Message Passing Networks for Molecular Dynamics Simulations

Geometry-Augmented Molecular Representation Learning for Property Prediction

Geometry-Based Molecular Generation with Deep Constrained Variational Autoencoder.

Self-Supervised Graph Information Bottleneck for Multi-View Molecular Embedding Learning

HAGO-Net: Hierarchical Geometric Massage Passing for Molecular Representation Learning

A Spatial-Temporal Gated Attention Module for Molecular Property Prediction Based on Molecular Geometry

Molecular geometric deep learning

Molecular CT: Unifying Geometry and Representation Learning for Molecules at Different Scales

3DMol-Net: Learn 3D Molecular Representation using Adaptive Graph Convolutional Network Based on Rotation Invariance

Self-Supervised Graph Information Bottleneck for Multiview Molecular Embedding Learning

Molecular Mechanics-Driven Graph Neural Network with Multiplex Graph for Molecular Structures