Molecular graph convolutions: moving beyond fingerprints

• Published in: Journal of computer-aided molecular design Vol. 30; no. 8; pp. 595 - 608
• Main Authors: Kearnes, Steven, McCloskey, Kevin, Berndl, Marc, Pande, Vijay, Riley, Patrick
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.08.2016; Springer Nature B.V
• Subjects: Animal Anatomy; Chemistry; Chemistry and Materials Science; Computer Applications in Chemistry; Computer Graphics; Computer-Aided Design; Convolution; Drug Design; Encoding; Fingerprinting; Fingerprints; Graphs; Histology; Learning; Ligands; Machine Learning; Machinery; Mathematical models; Molecular biology; Molecular Structure; Morphology; Neural Networks (Computer); Pharmaceutical Preparations - chemistry; Physical Chemistry; Deep learning; Virtual screening; Molecular descriptors; Machine learning; Artificial neural networks
• ISSN: 0920-654X; 1573-4951
• DOI: 10.1007/s10822-016-9938-8

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.