pytdc

To load drug discovery datasets for machine learning benchmarking, use the TDC dataset and task APIs to retrieve AI-ready therapeutic data with consistent schemas for modeling and assessment. You must install PyTDC via pip before calling the APIs to fetch ADME, toxicity, or interaction datasets.

A scaffold split partitions molecules based on their chemical structure to better measure generalization in drug discovery settings. You need scaffold splits for molecule property prediction when evaluating whether your model truly learns predictive rules rather than memorizing similar chemical scaffolds.

Yes, you can benchmark drug-target interaction models using standardized dataset splits. The framework provides multi-instance prediction datasets for DTI, DDI, and PPI tasks, offering consistent tabular representations and cold split strategies for training and assessment.

PyTDC supports both molecule generation tasks and property prediction workflows. It provides benchmark datasets and utilities for generation tasks alongside single-instance prediction for molecule and protein properties, covering a wide range of therapeutic machine learning applications.

You need to install PyTDC via pip and ensure SciPy is available in your environment to use these therapeutic ML datasets. SciPy is the required dependency for handling the standardized tabular representations and dataset loading operations.

Your drug discovery model evaluation may fail to generalize because random data splits leak structurally similar molecules into both training and test sets. Using scaffold splits or cold splits ensures structurally distinct compounds are separated, providing a rigorous measure of model generalization.