Geometric Graph Representations and Geometric Graph Convolutions for\n Deep Learning on Three-Dimensional (3D) Graphs

2020/06/02 by Daniel T. Chang, Chang, Daniel T.
Computer Science · #Advanced Graph Neural Networks #Computational Drug Discovery Methods #Computer Vision and Pattern Recognition (cs.CV) #Data Visualization and Analytics #FOS: Computer and information sciences #Graph Theory and Algorithms #Machine Learning (cs.LG)

paper · pdf · doi:10.48550/arxiv.2006.01785

openalex publication_date 2020/06/02 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

The geometry of three-dimensional (3D) graphs, consisting of nodes and edges,\nplays a crucial role in many important applications. An excellent example is\nmolecular graphs, whose geometry influences important properties of a molecule\nincluding its reactivity and biological activity. To facilitate the\nincorporation of geometry in deep learning on 3D graphs, we define three types\nof geometric graph representations: positional, angle-geometric and\ndistance-geometric. For proof of concept, we use the distance-geometric graph\nrepresentation for geometric graph convolutions. Further, to utilize standard\ngraph convolution networks, we employ a simple edge weight / edge distance\ncorrelation scheme, whose parameters can be fixed using reference values or\ndetermined through Bayesian hyperparameter optimization. The results of\ngeometric graph convolutions, for the ESOL and Freesol datasets, show\nsignificant improvement over those of standard graph convolutions. Our work\ndemonstrates the feasibility and promise of incorporating geometry, using the\ndistance-geometric graph representation, in deep learning on 3D graphs.\n

Citations

• Practical Bayesian Optimization of Machine Learning Algorithms
• Taking the Human Out of the Loop: A Review of Bayesian Optimization
• Bayesian Optimization for Machine Learning : A Practical Guidebook
• A Tutorial on Bayesian Optimization
• Fast Graph Representation Learning with PyTorch Geometric
• Detecting Functionality-Specific Vulnerabilities via Retrieving Individual Functionality-Equivalent APIs in Open-Source Repositories
• Tiered Graph Autoencoders with PyTorch Geometric for Molecular Graphs
• Bayesian Hyperparameter Optimization with BoTorch, GPyTorch and Ax

Related

• Distance-Geometric Graph Convolutional Network (DG-GCN) for Three-Dimensional (3D) Graphs
• Distance-Geometric Graph Attention Network (DG-GAT) for 3D Molecular Geometry
• Is Distance Matrix Enough for Geometric Deep Learning?
• Geometrically Principled Connections in Graph Neural Networks
• Graph representations of 3D data for machine learning