admet-prediction

ADMET Property Prediction

Overview

ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) profiling is one of the most critical stages in the drug discovery pipeline. Poor pharmacokinetic and toxicity properties are responsible for roughly 40% of clinical trial failures. Computational ADMET prediction enables medicinal chemists to triage compounds early, prioritize synthesis efforts, and design molecules with improved drug-like profiles before committing to expensive in vitro and in vivo studies.

This skill provides a comprehensive suite of ADMET prediction tools built on RDKit molecular descriptors, validated SMARTS-based structural alert libraries, and established empirical models. Every prediction is accompanied by a traffic-light classification (GREEN / YELLOW / RED) so that results can be interpreted at a glance.

When to Use This Skill

• Hit-to-lead optimization: Rank hits from a high-throughput screen by their predicted ADMET profile.
• Lead optimization: Identify liabilities in a lead series and guide structural modifications.
• Virtual screening triage: Filter large compound libraries before docking or ML scoring.
• Candidate comparison: Compare your candidates head-to-head against each other or against FDA-approved drug statistics.
• Toxicity flagging: Run focused toxicity panels before committing to synthesis.
• Drug-likeness gating: Evaluate whether a compound class is suitable for oral, CNS, topical, or injectable delivery.

Installation

Required dependencies

pip install rdkit-pypi numpy pandas

rdkit-pypi provides the pure-Python RDKit wheel. On conda-based environments, use conda install -c conda-forge rdkit instead.

Optional dependencies

pip install PyTDC deepchem

• PyTDC (Therapeutics Data Commons): provides access to benchmark ADMET datasets and pre-trained models for endpoints like Caco-2, hERG, CYP inhibition, and clearance.
• DeepChem: enables deep-learning-based ADMET models (e.g., graph neural networks for solubility and toxicity).