Directional Message Passing for Molecular Graphs (2020-03-06T00:00:00.000000Z)

TL;DR

This work proposes a message passing scheme analogous to belief propagation, which uses the directional information by transforming messages based on the angle between them, and uses spherical Bessel functions to construct a theoretically well-founded, orthogonal radial basis that achieves better performance than the currently prevalent Gaussian radial basis functions while using more than 4x fewer parameters.

Abstract

Graph neural networks have recently achieved great successes in predicting quantum mechanical properties of molecules. These models represent a molecule as a graph using only the distance between atoms (nodes) and not the spatial direction from one atom to another. However, directional information plays a central role in empirical potentials for molecules, e.g. in angular potentials. To alleviate this limitation we propose directional message passing, in which we embed the messages passed between atoms instead of the atoms themselves. Each message is associated with a direction in coordinate space. These directional message embeddings are rotationally equivariant since the associated directions rotate with the molecule. We propose a message passing scheme analogous to belief propagation, which uses the directional information by transforming messages based on the angle between them. Additionally, we use spherical Bessel functions to construct a theoretically well-founded, orthogonal radial basis that achieves better performance than the currently prevalent Gaussian radial basis functions while using more than 4x fewer parameters. We leverage these innovations to construct the directional message passing neural network (DimeNet). DimeNet outperforms previous GNNs on average by 77% on MD17 and by 41% on QM9.

Authors

Stephan Günnemann

9 papers

Johannes Klicpera

4 papers

Janek Groß

1 papers

TL;DR

Abstract

Authors

References47 items

Introduction to quantum mechanics, 3rd edition

Cormorant: Covariant Molecular Neural Networks

Provably Powerful Graph Networks

Generative Models for Graph-Based Protein Design

PhysNet: A Neural Network for Predicting Energies, Forces, Dipole Moments, and Partial Charges.

Gauge Equivariant Convolutional Networks and the Icosahedral CNN

Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals

Weisfeiler and Leman Go Neural: Higher-order Graph Neural Networks

How Powerful are Graph Neural Networks?

Predict then Propagate: Graph Neural Networks meet Personalized PageRank

Deep reinforcement learning with relational inductive biases

3D Steerable CNNs: Learning Rotationally Equivariant Features in Volumetric Data

Predicting molecular properties with covariant compositional networks.

Clebsch-Gordan Nets: a Fully Fourier Space Spherical Convolutional Neural Network

Neural Message Passing with Edge Updates for Predicting Properties of Molecules and Materials

RotDCF: Decomposition of Convolutional Filters for Rotation-Equivariant Deep Networks

Towards exact molecular dynamics simulations with machine-learned force fields

Tensor Field Networks: Rotation- and Translation-Equivariant Neural Networks for 3D Point Clouds

On the Convergence of Adam and Beyond

Spherical CNNs

Prediction Errors of Molecular Machine Learning Models Lower than Hybrid DFT Error.

SchNet: A continuous-filter convolutional neural network for modeling quantum interactions

Supervised Community Detection with Line Graph Neural Networks

Neural Message Passing for Quantum Chemistry

MoleculeNet: a benchmark for molecular machine learning

Equivariance Through Parameter-Sharing

Machine learning prediction errors better than DFT accuracy

Steerable CNNs

Machine learning of accurate energy-conserving molecular force fields

Semi-Supervised Classification with Graph Convolutional Networks

Group Equivariant Convolutional Networks

Deep Residual Learning for Image Recognition

Convolutional Networks on Graphs for Learning Molecular Fingerprints

Quantum chemistry structures and properties of 134 kilo molecules

On representing chemical environments

A Neural Device for Searching Direct Correlations Between Structures and Properties of Chemical Compounds.

Gaussian approximation potentials: the accuracy of quantum mechanics, without the electrons.

Simultaneous fitting of a potential-energy surface and its corresponding force fields using feedforward neural networks.

Generalized neural-network representation of high-dimensional potential-energy surfaces.

A new model for learning in graph domains

A Neural Device for Searching Direct Correlations between Structures and Properties of Chemical Compounds

Supervised neural networks for the classification of structures

Searching for Activation Functions

Supplementary Materials for Machine learning unifies the modeling of materials and molecules

The Graph Neural Network Model

Understanding belief propagation and its generalizations

Molecular modelling: Principles and applications

Field of Study

Journal Information

Name

Volume

Venue Information

Name

Type

URL

Alternate Names