A Comparative Study of QSPR Methods on a Unique Multitask PAMPA Data Set

We present a unique, multitask data set comprising 143 drug and drug candidate molecules, each evaluated on in vitro parallel artificial membrane permeability assays (PAMPA) using six different model membranes. Using this resource, we systematically assessed the effectiveness of various molecular descriptors and regression models in predicting passive membrane permeability. The studied models range from simple linear regression to a modern pretrained transformer architecture. Particular attention is given to the trade-off between predictive performance and model interpretability, highlighting the challenges introduced by machine learning approaches. To our knowledge, this is the most comprehensive study of simultaneous modeling of multiple organ-specific PAMPA membranes to date, offering novel insights into membrane-specific permeability profiles. We found that expert-designed physicochemical property descriptors are more fitting for a limited sample size permeability study than deep learning-based representations.