Molecular graph convolutions: moving beyond fingerprints

By Steven M. Kearnes, Kevin McCloskey, Marc Berndl, Vijay S. Pande, Patrick F. Riley. 2016

1,608 citations recorded by the source corpus.

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.

Subfields: Biomedical Engineering, Computational Theory and Mathematics, Materials Chemistry, Graph Learning, Computational Chemistry & Drug Discovery

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Biomedical Engineering

In Computational Theory and Mathematics

In Materials Chemistry

In Graph Learning

In Computational Chemistry & Drug Discovery

Citations in the corpus

Axiomatic Attribution for Deep Networks (2017, methodology): “Figure 6: Attribution for a molecule under the W2N2 network (Kearnes et al., 2016) .”
Neural Message Passing for Quantum Chemistry (2017, methodology): “Molecular Graph Convolutions, Kearnes et al. (2016) This work deviates slightly from other MPNNs in that it introduces edge representations etvw which are updated during the message passing phase.”
Graph Convolutional Neural Networks for Web-Scale Recommender Systems (2018, methodology): “, node2vec [17] and DeepWalk [26]), and their success has led to a surge of interest in applying GCN-based methods to applications ranging from recommender systems [24] to drug design [20, 31].”
Relational inductive biases, deep learning, and graph networks (2018, positive): “Kearnes et al. (2016) computes edge updates from nodes in a similar manner.”
Unsupervised Inductive Whole-Graph Embedding by Preserving Graph Proximity (2019, methodology): “• Others: Other methods include Capsule Graph Neural Network (Zhang & Chen, 2019),aggregation using “fuzzy” histograms (Kearnes et al., 2016), Convolutional Neural Networks (Niepert et al.”
Graph Warp Module: an Auxiliary Module for Boosting the Power of Graph Neural Networks (2019, background): “Neural Fingerprint (NFP) [5] and Weavenet [10] are relatively classical architectures.”
How Powerful are Graph Neural Networks? (2018, background): “…schemes have been proposed (Scarselli et al., 2009b; Battaglia et al., 2016; Defferrard et al., 2016; Duvenaud et al., 2015; Hamilton et al., 2017a; Kearnes et al., 2016; Kipf & Welling, 2017; Li et al., 2016; Velickovic et al., 2018; Santoro et al., 2017; Xu et al., 2018; Santoro et al., 2018;…”
Representation Learning on Graphs with Jumping Knowledge Networks (2018, background): “Many of these approaches broadly follow a neighborhood aggregation (or “message passing” scheme), and those have been very promising (Kipf & Welling, 2017; Hamilton et al., 2017; Gilmer et al., 2017; Veliˇckovi´c et al., 2018; Kearnes et al., 2016).”
Graph Convolutional Policy Network for Goal-Directed Molecular Graph Generation (2018, methodology): “In order to perform link prediction in G t ∪ C , our model ﬁrst computes the node embeddings of an input graph using Graph Convolutional Networks (GCN) [20, 5, 18, 36, 8], a well-studied technique that achieves state-of-the-art performance in representation learning for molecules.”
Mol2vec: Unsupervised Machine Learning Approach with Chemical Intuition (2018, neutral): “8 Most worth noting, the NLP method “ term frequency-inverse document frequency” (tf-idf) was applied on Morgan fingerprints for compound-protein prediction 11 and the ” Latent Dirichlet Allocation” method for chemical topic modeling.”
Representation Learning on Graphs: Methods and Applications (2017, background): “Subgraph embeddings can be used to classify or predict various properties of molecular graphs, including predicting the efficacy of potential solar cell materials [16], or predicting the therapeutic effect of candidate drugs [33].”
Opportunities and obstacles for deep learning in biology and medicine (2017, methodology): “In the original study, graph convolutions [310] performed comparably to a multi-task network using standard fingerprints and slightly better than the neural fingerprints [308] on the drug efficacy task but were slightly worse than the influence-relevance voter method on an HIV dataset.”
Ligand biological activity predicted by cleaning positive and negative chemical correlations (2019, background): “Pioneering advances in machine learning that infer the optimal representation of molecules directly from data (23–26) do not resolve this undersampling problem, as the available data are usually signiﬁcantly less than the number of parameters in the model.”
Learning Multimodal Graph-to-Graph Translation for Molecular Optimization (2018, methodology): “, 2017) and chemistry (Kearnes et al., 2016; Gilmer et al., 2017; Schütt et al., 2017; Jin et al., 2017).”
Junction Tree Variational Autoencoder for Molecular Graph Generation (2018, methodology): “For applications, graph neural networks are used in semisupervised classification (Kipf & Welling, 2016), computer vision (Monti et al., 2016), and chemical domains (Kearnes et al., 2016; Schütt et al., 2017; Jin et al., 2017).”