One multitask model predicts the full ADMET panel, with a focus on natural products
- hdADMET is a single multitask model that links compounds, known drugs, measured ADMET data and relationships between endpoints in a network, then feeds the learned embeddings into a kernel-based deep neural network.
- It was tested with hold-out splits on public bioactivity and natural-product datasets; no prospective experimental validation is reported.
- Performance was competitive with leading multitask baselines at much lower compute, and it profiled newly literature-mined natural products, flagging one candidate for follow-up.
💡 Points Based on the abstract
"Competitive at lower compute" only means something if the hold-out splits were scaffold- or cluster-based; random splits on public ADMET sets reward near-neighbour lookup, and a network that links compounds to known drugs is especially prone to that leakage. Before trusting it on our series, I'd run it against a chemprop or similar multitask baseline on our own time-split internal data, and check calibration on natural-product chemotypes that sit far from the training space.
📄 Paper · Journal of Chemical Information and Modeling (ACS) ↗ Large-Scale and Systematic Prediction of ADMET Properties through Heterogeneous Network and Kernel-Based Deep Learning