name: 'chemcrow-drug-discovery' description: 'An LLM chemistry agent with expert-designed tools for organic synthesis, drug discovery, and materials design.' measurable_outcome: Execute skill workflow successfully with valid output within 15 minutes. allowed-tools:
- read_file
- run_shell_command
ChemCrow is an open-source package for the accurate integration of Large Language Models (LLMs) with chemistry tools. It is designed to autonomously plan and execute chemical syntheses, research materials, and discover new drugs.
- Synthesis Planning: "How do I synthesize Ibuprofen?"
- Property Prediction: "What is the logP of this molecule?"
- Safety Checks: "Is this reaction explosive?"
- Literature Search: "Find patents related to this substructure."
- Synthesis Planning: Uses tools like RXN4Chemistry or local retrosynthesis models to plan routes.
- Molecule Manipulation: Add/remove functional groups, generate SMILES (RDKit).
- Safety Assessment: Checks reagents against safety databases (PubChem, GHS).
- Web Search: Google/Patent search for chemical literature.
- RDKit: Cheminformatics (MolToSmiles, SmilesToMol, descriptors).
- PubChem: Search for properties and safety data.
- LiteratureSearch: Search arXiv, patents.
- Synthesis: IBM RXN, OPLS.
- Name2SMILES: Convert chemical names to structures.
User: "Plan a synthesis for a derivative of Aspirin that might have better solubility."
Agent Action:
- Retrieves Aspirin structure (SMILES).
- Modifies structure (e.g., adds a polar group) using RDKit.
- Checks solubility prediction.
- Plans synthesis route for the new molecule.
- Checks safety of reagents.