Pytdc

If you're doing drug discovery ML, this gives you standardized datasets and benchmarks for the entire therapeutics pipeline. It covers single-instance predictions like ADME and toxicity, multi-instance tasks like drug-target interactions, and generation workflows for novel molecules. The nice part is the scaffold splits and cold-split methods that actually test generalization, plus molecular oracles for property optimization. It's from the Therapeutics Data Commons project and wraps everything in a consistent API so you're not hunting down disparate datasets. Worth using if you're benchmarking models on pharmaceutical tasks or need curated train/test splits that reflect real-world drug discovery challenges rather than random shuffles.