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

Abstract

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

arXiv commentChemical Science, in press

References in corpus (9)

• Neural Message Passing for Quantum Chemistry
• Automatic chemical design using a data-driven continuous representation of molecules
• Convolutional Networks on Graphs for Learning Molecular Fingerprints
• Deep Bayesian Active Learning with Image Data
• Most Ligand-Based Classification Benchmarks Reward Memorization Rather than Generalization
• Using Attribution to Decode Dataset Bias in Neural Network Models for Chemistry
• Dropout as a Bayesian Approximation: Appendix
• On the Importance of Strong Baselines in Bayesian Deep Learning
• Semi-supervised learning of hierarchical representations of molecules using neural message passing

Cited by in corpus (17)

• A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges
• Predicting materials properties without crystal structure: Deep representation learning from stoichiometry
• Exploring Chemical Space using Natural Language Processing Methodologies for Drug Discovery
• Reliable Prediction Errors for Deep Neural Networks Using Test-Time Dropout
• Constrained Bayesian Optimization for Automatic Chemical Design
• On Statistical Bias In Active Learning: How and When To Fix It
• Gaussian Process Molecule Property Prediction with FlowMO
• Artificial Intelligence for Science in Quantum, Atomistic, and Continuum Systems
• Quantifying sources of uncertainty in drug discovery predictions with probabilistic models
• Graph Posterior Network: Bayesian Predictive Uncertainty for Node Classification
• Evaluating Scalable Uncertainty Estimation Methods for DNN-Based Molecular Property Prediction
• Ranking over Regression for Bayesian Optimization and Molecule Selection
• A comprehensive study on the prediction reliability of graph neural networks for virtual screening
• Mitigating the Effects of Non-Identifiability on Inference for Bayesian Neural Networks with Latent Variables
• Entropy-based Active Learning of Graph Neural Network Surrogate Models for Materials Properties
• GNN's Uncertainty Quantification using Self-Distillation
• Uncertainty-Aware Robust Learning on Noisy Graphs