tooluniverse-chemical-safety

Provide the SMILES string and the Skill runs ADMET-AI predictions for AMES mutagenicity, DILI, hERG inhibition, LD50, carcinogenicity, and skin reaction, plus full ADMET property profiling. Results are graded as computational [T3] evidence and combined into a risk classification.

It integrates six sources: ADMET-AI for predictive toxicology, CTD for chemical-gene-disease toxicogenomics, FDA labels for regulatory safety, DrugBank for drug safety profiles, STITCH for chemical-protein interactions, and ChEMBL for structural alerts, with PubChem for compound disambiguation.

Yes. For environmental chemicals, the Skill relies on ADMET-AI predictions, CTD chemical-disease associations, and STITCH interactions, while FDA and DrugBank phases report 'not applicable'. CTD coverage of contaminants like bisphenol A makes it suitable for environmental toxicology.

Every finding is tagged [T1] through [T4]: T1 is direct human or regulatory evidence like FDA boxed warnings, T2 is animal or validated in vitro data, T3 is computational prediction, and T4 is unvalidated database annotation. The final risk classification cites these tiers.

ADMET-AI predictions are computational [T3] evidence and should not replace experimental testing. Invalid SMILES cause prediction failures, and novel compounds may have no supporting database evidence, leaving predictions as the only data source.

Yes. ADMET-AI tools accept a list of up to roughly 10 SMILES per call, enabling batch toxicity and ADMET prediction. The Skill supports a batch screening pattern that produces a comparative table followed by a synthesized risk summary.