Equivariant Graph Attention Networks for Molecular Property Prediction

TL;DR

This paper introduces an equivariant graph attention network that uses Cartesian coordinates and a novel attention mechanism to improve molecular property prediction, especially for quantum and macromolecular structures.

Contribution

It proposes a new equivariant GNN with a content and spatial dependent attention mechanism for better molecular structure modeling.

Findings

Improved accuracy in quantum mechanical property prediction.

Enhanced modeling of protein complexes.

Effective incorporation of directionality in molecular graphs.

Abstract

Learning and reasoning about 3D molecular structures with varying size is an emerging and important challenge in machine learning and especially in drug discovery. Equivariant Graph Neural Networks (GNNs) can simultaneously leverage the geometric and relational detail of the problem domain and are known to learn expressive representations through the propagation of information between nodes leveraging higher-order representations to faithfully express the geometry of the data, such as directionality in their intermediate layers. In this work, we propose an equivariant GNN that operates with Cartesian coordinates to incorporate directionality and we implement a novel attention mechanism, acting as a content and spatial dependent filter when propagating information between nodes. We demonstrate the efficacy of our architecture on predicting quantum mechanical properties of small molecules and its benefit on problems that concern macromolecular structures such as protein complexes.