Improvement in ADMET predictions continues to be of great interest in drug discovery and to those performing drug design. Here, we report systematic application of random forest and deep learning approaches to prediction of ADMET endpoints. We report a step-change improvement in computational ADMET prediction with the PotentialNet family of graph convolutional approaches, described in a previous talk. Traditional cheminformatics approaches, whether the learner is a random forest or a deep neural network, leverage fixed fingerprint feature representations of molecules. In contrast, we "learn" the features most relevant to each chemical task at hand by representing each molecule explicitly as a graph, where each node is an atom and each edge is a bond. Using these explicit molecular representations, we achieve unprecedented accuracy in prediction of ADMET endpoints, with remarkable improvement in a subset of endpoints with common characteristics. We show that deep featurization better enables molecular predictors to not only interpolate, but to extrapolate to new regions of chemical space. Finally, we validate our model predictions with prospective experiments.

Comparison of random forest and deep learning approaches for ADMET endpoint prediction

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.