MoleculeNet

MoleculeNet is designed as a benchmark for machine learning methods focused on molecular properties, aiming to advance molecular machine learning. Numerous datasets have been collected and organized, and they are provided in combination with software suites that implement a variety of featurization techniques and past algorithms. These resources are integrated as part of the open-source DeepChem package, which includes data on over 700,000 compounds and their various properties. Various machine learning models are tested on these datasets, and the results are reported in AUC-ROC, AUC-PRC, RMSE, and MAE scores.