Reproduce “Reliable Graph Neural Networks for Drug Discovery Under Distributional Shift” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kehang Han, Balaji Lakshminarayanan, Jeremiah Liu
GNN-SNGP adapts a graph neural network to estimate molecular cardiotoxicity with uncertainty tied to distance from training data. It combines spectral normalization and skip connections in message passing with a Gaussian-process classifier. On the FDA-sourced Test-OOD2, it achieved AUROC 0.885±0.007 and ECE 0.044±0.006, versus 0.831±0.007 and 0.082±0.012 for the baseline, reducing but not eliminating overconfident false negatives. These results concern one benchmark setting, not clinical validation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.