pharmaclaw-tox-agent

What This Skill Does

The pharmaclaw-tox-agent is a specialized AI agent within the OpenClaw ecosystem designed to perform rapid, descriptor-based toxicology and drug safety profiling directly from SMILES strings. By leveraging the RDKit library, this tool quantifies essential medicinal chemistry parameters including Molecular Weight (MW), lipophilicity (logP), topological polar surface area (TPSA), and hydrogen bond donor/acceptor counts. It automates critical safety assessments by enforcing Lipinski’s Rule of Five and Veber’s rules for drug-likeness. Furthermore, it calculates the Qualitative Estimate of Drug-likeness (QED) score and screens for PAINS (Pan-Assay Interference Compounds) substructures to detect potential false positives in high-throughput screening assays. The agent provides a structured risk classification, allowing researchers to triage candidates effectively, and feeds directly into downstream workflows like IP expansion for safer derivative synthesis.

Installation

To integrate this skill into your OpenClaw environment, use the internal package manager. Ensure your environment has the necessary chemistry stack installed. Run the following command in your terminal:

clawhub install openclaw/skills/skills/cheminem/pharmaclaw-tox-agent

Use Cases

This skill is indispensable for early-stage drug discovery projects where chemical space exploration needs to be balanced with safety. It is primarily used during the hit-to-lead and lead optimization phases. Researchers utilize this to:

• Rapidly filter large virtual libraries for acceptable pharmacokinetic profiles.
• Assess the safety profile of novel scaffold derivatives before proceeding to synthesis.
• Identify PAINS alerts early to prevent costly, futile lab experiments.
• Rank candidate compounds by their QED scores to prioritize the most drug-like molecules for subsequent pharmacology testing.

Example Prompts

1. "Run a toxicology safety profile for the following SMILES: CC(=O)Nc1ccc(O)cc1 and tell me if it passes the Lipinski Rule of Five."
2. "Analyze this compound for PAINS alerts and check the TPSA: [SMILES_STRING]"
3. "Evaluate the drug-likeness and risk profile of my recent derivative and suggest if it is safe for further synthesis."

Tips & Limitations

The pharmaclaw-tox-agent is intended as an early-stage screening tool. While it provides high-speed automated evaluation, it relies on descriptor-based proxies rather than deep-learning models for specific toxicity endpoints like Ames mutagenicity or hERG channel inhibition. Users should be aware that the current PAINS implementation uses a simplified substructure set; for final regulatory-grade submissions, comprehensive validation using the full PAINS catalog is recommended. Always combine these automated metrics with expert medicinal chemistry intuition.