Computational & AI chemRxiv

Lipid composition and permeant properties jointly shape passive membrane permeation free energies in MD

  • Adaptive-weight-histogram all-atom MD (MemCross) computed permeation free energy profiles for a small, structurally varied set of drug-like molecules.
  • Bilayers were POPC mixed in turn with cholesterol, POPE, a polyunsaturated PC, or sphingomyelin, mimicking plasma membrane diversity.
  • The work is purely computational; permeation depended on both lipid makeup and permeant properties, and mechanistic descriptors were distilled.
  • Multivariate models built on these descriptors point to quantities obtainable from cheaper unbiased simulations.
Points Based on the abstract

Six compounds can't support multivariate models, so treat the descriptors as hypotheses, not a predictor. Without a PAMPA or Caco-2 anchor, we also can't tell whether lipid-specific free-energy shifts show up in assays, and transporters and efflux are ignored. Still useful: if unbiased-MD descriptors reproduce these shifts, they could add physics-based features to ML permeability models.