Impact of Hydrogen Bonding on P‑Glycoprotein Efflux Transport as Revealed by Evaluation of a De Novo Prediction Model

Drug efflux transport by the permeability glycoprotein (P-gp) frequently diminishes drug efficacy, making the prediction of the P-gp efflux ratio critically significant. Most contemporary computational methods rely on binary predictions through machine learning. However, the accuracy and applicability of these predictions are heavily influenced by the training data sets. In contrast, we evaluated the validity of a de novo prediction model, which employs solvation free energies both within and outside the P-gp-binding pocket. Utilizing our in-house data set of 397 compounds, we discovered that this model struggles to predict a certain class of compounds accurately and that such outliers often exhibit favorable solvation free energy, with an increased number of hydrogen bond donors compared to other compounds investigated. Further examination of the functional groups in these compounds highlighted the significance of their specific interactions with P-gp. Considering these specific drug-P-gp interactions could enhance the accuracy of mechanism-based de novo prediction methods.