MPCD: Benchmark datasets of molecular property cliff (MPC) in ACANet paper.

MPCD This release provides the Molecular Property Cliff Dataset (MPCD) associated with the manuscript: Activity-cliff awareness enables robust graph learning for predicting molecular properties MPCD is a benchmark dataset collection designed for evaluating molecular machine learning models under activity-cliff-aware settings. The released repository provides curated molecular property cliff datasets used for model training, evaluation, and comparative analysis in the ACANet study. Release contents This release includes benchmark datasets covering four major evaluation settings: Low-sample-size narrow-scaffold datasets (LSSNS) Nine activity prediction datasets with limited sample sizes and narrow scaffold distributions, designed to evaluate model robustness in small-data and scaffold-constrained scenarios. High-sample-size mixed-scaffold datasets (HSSMS) Thirty molecular activity prediction datasets derived from activity-cliff benchmark tasks, covering diverse targets, target families, assay types, and scaffold distributions. Matched molecular pair datasets (MMP) Three matched molecular pair datasets for activity-cliff classification, supporting the evaluation of models on structurally similar compound pairs with large activity differences. ADMET delta prediction datasets Ten ADMET-related datasets for delta prediction tasks, covering absorption, distribution, metabolism, excretion, and toxicity properties. License This repository is released under the Apache-2.0 license. Citation If you use MPCD in your work, please cite the associated manuscript and this archived software/data release.