Logo: to the web site of Uppsala University

With the current move from classical small molecules to new synthetic modalities, there is a strong need to predict the pharmacokinetics of beyond and extended rule of five (bRo5 & eRo5) compounds. The present master’s thesis introduces two machine learning models for classification and regression of Caco-2 permeability of compounds belonging to these bRo5 and ebRo5 categories, to support scientist in selection of compounds during the early drug discovery process. Different algorithms and descriptors are explored and the final regression and classification model evaluated. Additionally, an approach is developed to predict human oral exposures from key parameters such as clearance and permeability. Different clearance models are included and assessed by comparing resulting exposure projections to literature values, which revealed fundamental challenges related to reliable estimation of exposure. The presented permeability classification model can be used to categorize compounds according to three defined permeability classes and can therefore be a valuable tool for medicinal chemists in a drug discovery process, while the regression model can be helpful to avoid synthesis efforts of low permeable compounds.