Bayesian semi-supervised learning for uncertainty-calibrated prediction of molecular properties and active learning.

• DOI: 10.17863/cam.48436
• Publication Date: 2019-07-10
• Authors: Yao Zhang; Alpha A. Lee
• Keywords: Representation

Predicting bioactivity and physical properties of small molecules is a central challenge in drug discovery. Deep learning becoming the method choice but studies to date focus on mean accuracy as main metric. However, replace costly mission-critical experiments by models, high not enough: outliers can derail discovery campaign, thus models need reliably predict when it will fail, even training data biased; are expensive, be data-efficient suggest informative sets using active learning. We show that uncertainty quantification achieved Bayesian semi-supervised graph convolutional neural networks. The approach estimates statistically principled way through sampling from posterior distribution. Semi-supervised disentangles representation regression, keeping accurate low limit allowing model start initial pool data. Our study highlights promise deep for chemistry.