Distance-Geometric Graph Convolutional Network (DG-GCN) for Three-Dimensional (3D) Graphs

TL;DR

This paper introduces DG-GCN, a novel graph convolutional network that leverages distance-geometric representations to effectively incorporate 3D geometry into deep learning models, improving performance on molecular graph datasets.

Contribution

It proposes a new DG-GCN model that uses continuous-filter convolutional layers with distance-based filters, enhancing 3D graph learning capabilities.

Findings

Major performance improvements on ESOL and FreeSolv datasets.

Outperforms standard and geometric graph convolutions.

Demonstrates effectiveness for molecular graph analysis.

Abstract

The distance-geometric graph representation adopts a unified scheme (distance) for representing the geometry of three-dimensional(3D) graphs. It is invariant to rotation and translation of the graph and it reflects pair-wise node interactions and their generally local nature. To facilitate the incorporation of geometry in deep learning on 3D graphs, we propose a message-passing graph convolutional network based on the distance-geometric graph representation: DG-GCN (distance-geometric graph convolution network). It utilizes continuous-filter convolutional layers, with filter-generating networks, that enable learning of filter weights from distances, thereby incorporating the geometry of 3D graphs in graph convolutions. Our results for the ESOL and FreeSolv datasets show major improvement over those of standard graph convolutions. They also show significant improvement over those of geometric graph convolutions employing edge weight / edge distance power laws. Our work demonstrates the utility and value of DG-GCN for end-to-end deep learning on 3D graphs, particularly molecular graphs.