Membrane free-energy profiles from SILCS feed ML permeability models
- Free-energy profiles across lipid bilayers, computed with SILCS for several thousand compounds and accounting for protonation state, serve as input features for machine-learning permeability models.
- Three membrane models are used: a cholesterol-containing bilayer, a PAMPA-like membrane and a blood-brain barrier model.
- Retrospective evaluation only; accuracy is reported as comparable to published models, with per-functional-group energy contributions to guide optimization. Features based on functional-group energetics may generalize better to new chemistry; cell-assay transporter effects limit the training data.
💡 Points Based on the abstract
"Comparable to published models" is the number to interrogate: if a fingerprint model on the same cell-assay data does equally well, the simulation cost buys only interpretability. That per-group energy attribution is the real appeal, but it needs a scaffold-split or time-split test and a held-out series from your own project before you trust it. Transporter-confounded cell-assay labels will cap any passive-diffusion physics.
📄 Paper · Journal of Chemical Information and Modeling (ACS) ↗ Learning from SILCS FragMaps: Combining Physics-Based Energies with Artificial Intelligence for Rapid Drug Permeability Prediction