LipinskiFix-1000 dataset

The Lipinski-1000 dataset proposed by "AutoLead: An LLM-Guided Bayesian Optimization Framework for Multi-Objective Lead Optimization", a benchmark dataset designed to evaluate molecular optimization strategies on realistic drug-like compounds. It consists of 1,000 unique ligands extracted from the high-quality HiQBind~\cite{wang2025workflow} database, each bound to experimentally validated protein targets. These molecules were selected for violating at least one of Lipinski's Rule of Five criteria, reflecting common challenges in early drug discovery. To ensure meaningful optimization, we applied additional quality constraints on molecular weight, lipophilicity, and QED scores, while limiting structural complexity. Each task targets restoring Lipinski compliance as the primary objective and maximizing QED as a secondary goal, providing a rigorous testbed for multi-objective molecular design.