One model scores peptide PK and toxicity from SMILES or sequence
- Accepts either SMILES or amino acid sequences, so natural, modified, cyclic and peptide-like inputs are covered; predicts permeability (Caco-2, PAMPA), half-life, clearance, distribution volume, neurotoxicity and hemolysis.
- Trained on data compiled from public databases and literature; several encodings and model families, from classical ML to transformers and language models, were compared.
- Adds a function-aware score judging suitability given a peptide's intended activity. Reported good accuracy and interpretability on independent test sets (retrospective, no experimental validation described in the abstract).
💡 Points Based on the abstract
Peptide ADME data is small, noisy and scattered across assay protocols, so a single model spanning Caco-2, PAMPA, half-life and clearance deserves a hard look at split design: random splits on near-identical sequences inflate scores. Ask whether the test sets were clustered by sequence or scaffold, and whether it beats a simple baseline on your own cyclic or modified series before trusting any number.
📄 Paper · PLOS Computational Biology (PLOS) ↗ Pepti-DrugPhAI: Multimodal AI-evaluator for functional pharmacokinetics and toxicity of peptides