A fast and reliable approach for predicting membrane permeation of drugs was introduced. An advanced implicit membrane model, Heterogeneous Dielectric Generalized Born (HDGB), was used to account for the inhomogeneous character of the membrane environment. In addition to the standard HDGB, dynamic HDGB (DHDGB) was also used to allow for the membrane deformation upon the insertion of a drug molecule. The (D)HDGB models were used to obtain the free energy profiles of membrane partitioning of drug molecules and those profiles are then used to predict the permeability coefficients using solubility diffusion theory and machine learning separately. Neutralization of charged molecules upon the insertion of the drugs to the membrane was taken into account by using a hybrid free energy profiles obtained from neutral and charged molecules and the energetic cost of the charge neutralization. Three widely used force fields, CGenFF, OPLS and AM1-BCC were used to compare the predictions from the different partial charge sets. Theoretical estimations of permeability coefficients using both physics-based and machine learning approaches give good agreement with the experimental PAMPA permeabilities. The highest correlations were obtained using solubility diffusion theory based on the hybrid free energy profiles obtained by the DHDGB model and OPLS partial charge set. Machine learning provided comparable results but could not outperform the physics-based calculations.<br/>

Estimation of membrane permeation using implicit membrane models and machine learning

2019 Fall National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.