Molecular Graph Convolutions: Moving Beyond Fingerprints

TL;DR

This paper introduces molecular graph convolutions, a new machine learning approach that directly learns from molecular graph structures, offering a flexible alternative to traditional fingerprint methods in drug discovery.

Contribution

The paper presents molecular graph convolutions as a novel architecture for learning from molecular graphs, moving beyond fixed fingerprint representations.

Findings

Graph convolutions encode structural information effectively.

They offer a new paradigm for ligand-based virtual screening.

Performance is competitive but not yet superior to fingerprint methods.

Abstract

Molecular "fingerprints" encoding structural information are the workhorse of cheminformatics and machine learning in drug discovery applications. However, fingerprint representations necessarily emphasize particular aspects of the molecular structure while ignoring others, rather than allowing the model to make data-driven decisions. We describe molecular "graph convolutions", a machine learning architecture for learning from undirected graphs, specifically small molecules. Graph convolutions use a simple encoding of the molecular graph---atoms, bonds, distances, etc.---which allows the model to take greater advantage of information in the graph structure. Although graph convolutions do not outperform all fingerprint-based methods, they (along with other graph-based methods) represent a new paradigm in ligand-based virtual screening with exciting opportunities for future improvement.