molclaw-rgroup-optimization

What problem does it solve?

This Skill finds R-group decorations for a fixed scaffold and optimizes generated molecules toward multi-objective property targets (e.g., QED, molecular weight, LogP, TPSA) using reinforcement learning, eliminating manual trial-and-error and post-filtering inefficiencies.

Core Features & Use Cases

• RL-guided staged_learning combines LibInvent scaffold-constrained generation with REINVENT4's staged_learning optimization to produce property-aware decorated molecules.
• Multi-objective scoring and constraints supports weighted QED, MW, LogP, TPSA scoring, optional Tanimoto similarity to a reference, and R-group-specific constraints such as R-group MW and ring-count limits.
• Practical outputs and requirements expects a .smi scaffold file with [*:N] attachment points, validates file existence, and saves optimized molecules to a CSV with an uppercase SMILES column for downstream analysis; suitable for scaffold-constrained lead optimization and hit refinement.

Quick Start

Run libinvent_rgroup_optimization on a scaffolds.smi file to optimize R-groups for QED and property constraints using staged_learning with your chosen weights and max_steps.