This is a structured workflow for finding small molecule binders against protein targets using 60+ ToolUniverse tools. It walks Claude through druggability assessment, known ligand mining from ChEMBL and BindingDB, similarity expansion, ADMET filtering, and synthesis feasibility checks. The workflow is opinionated about methodology: it creates a markdown report file first, then populates it progressively while hiding tool outputs from the user. It includes binding site reasoning upfront to catch undruggable targets like flat protein-protein interfaces before wasting time on screens. Good for hit identification and virtual screening when you want a systematic, evidence-graded approach rather than ad-hoc compound searches. Requires NVIDIA API key for structure prediction via AlphaFold2 or ESMFold.
npx skills add https://github.com/mims-harvard/tooluniverse --skill tooluniverse-binder-discovery