admet-prediction

Goal

Compute ADMET-relevant physicochemical descriptors and rule-based drug-likeness heuristics from SMILES strings using RDKit.

This skill reports:

• Core descriptors: molecular weight (average and exact), Wildman-Crippen cLogP, TPSA, HBD/HBA, rotatable bonds, ring counts, aromatic rings, heavy atoms, fractionCSP3, molar refractivity.
• Heuristics:
  • Lipinski Rule of Five (Ro5) compliance (≤ 1 violation) as a permeability/absorption triage heuristic.
  • Veber oral bioavailability heuristic (RB ≤ 10 and TPSA ≤ 140 Ų; plus reporting the alternative HBD+HBA ≤ 12 condition).
  • QED (Quantitative Estimate of Drug-likeness) score.

Note: This does not predict experimental ADMET endpoints (e.g., clearance, CYP inhibition, hERG, Ames, etc.). It is an early-stage physchem/heuristics screen.

Instructions

The drugdisc MCP server provides a compute_molecular_descriptors tool that can be called directly: