Bayesian Graph Neural Networks for Molecular Property Prediction
This research provides an incremental improvement for chemists and material scientists using GNNs for molecular property prediction, particularly for generalization to new molecular scaffolds.
This study addresses the underspecification and generalization issues of graph neural networks (GNNs) in molecular property prediction by applying Bayesian learning. It benchmarks various Bayesian methods on a directed MPNN using the QM9 dataset, demonstrating that capturing uncertainty in both readout and message passing parameters improves predictive accuracy, calibration, and performance in molecular search.
Graph neural networks for molecular property prediction are frequently underspecified by data and fail to generalise to new scaffolds at test time. A potential solution is Bayesian learning, which can capture our uncertainty in the model parameters. This study benchmarks a set of Bayesian methods applied to a directed MPNN, using the QM9 regression dataset. We find that capturing uncertainty in both readout and message passing parameters yields enhanced predictive accuracy, calibration, and performance on a downstream molecular search task.