ImDrug: A Deep Imbalanced Learning Benchmark for AI-Aided Drug Discovery - a conversation

enjoy this great paper as a easy to understand conversation Summary The paper introduces ImDrug, a benchmark for evaluating deep imbalanced learning methods in AI-aided drug discovery. ImDrug addresses the prevalent issue of imbalanced datasets in this field, offering 11 datasets, 54 tasks, and 16 baseline algorithms. It features novel evaluation metrics (balanced accuracy and balanced F1) to mitigate biases from imbalanced data splits. The authors conduct extensive experiments across various imbalanced learning settings (classification and regression), highlighting the need for improved algorithms in this crucial area. ImDrug is open-source and provides tools for researchers to customize and expand the benchmark.