MolScore

MolScore unifies scoring functions, benchmarks, and evaluation metrics to facilitate generative molecular design and evaluation in de novo drug design.

Key Features:

• Scoring Function Integration: MolScore incorporates drug-design-relevant scoring functions including molecular similarity, molecular docking, predictive models, and synthesizability.
• Benchmark Unification: MolScore re-implements commonly used benchmarks in the field such as GuacaMol, MOSES, and MolOpt.
• Custom Benchmark Creation: New benchmarks can be created trivially through the platform's unified framework.
• Performance Metrics: MolScore provides performance metrics to evaluate generative model performance based on the chemistry generated.
• Python Integration: MolScore can be integrated into an existing Python script with three lines of code.

Scientific Applications:

• 5-HT2a Ligand Design: MolScore was demonstrated by testing a chemical language model with reinforcement learning on tasks related to the design of 5-HT2a ligands using molecular descriptors, 266 pre-trained QSAR models, or dual molecular docking.
• Generative Model Evaluation: MolScore enables evaluation of generative models for de novo drug design through unified scoring and benchmarking.

Methodology:

MolScore evaluates generative models by applying scoring functions to generated chemistry and computing performance metrics; specific computational algorithms or pipeline steps are not detailed in the sources.