Abstract

ABC-transporters such as P-glycoprotein (P-gp), the breast cancer resistance protein (BCRP) or the bile salt export pump (BSEP) play an important role in ADME, drug-drug interactions and drug-induced liver injury. Predicting their inhibition by small molecules may allow detecting drug-drug interaction perpetrators or prevent liver injury. The quest for a multidrug resistance reverting drug has led to the publication of many medicinal chemistry studies around P-gp and BCRP, bringing crucial data to build predictive models. Similarly, the publication and refinement of the crystal structure of Mouse P-gp has helped building three-dimensional models of human ABC-transporters. This thesis focuses on predicting and understanding the inhibition of liver ABC- transporters by small molecules using computational methods. First, literature data was collected and assembled to build a dataset around BCRP inhibition. Then, machine learning models were built to predict inhibition of P-gp, BCRP and BSEP by small molecules. They were used as virtual screening tools to successfully discover new inhibitors of BSEP and BCRP among small molecules from the DrugBank database. Two studies focused on understanding the specificity and overlap of inhibitors between P-gp and BCRP. One brought known multilabel learning methods to the chemoinformatics field while the other focused on propafenone analogues. Finally, a homology model of the transmembrane domain of BCRP was built and used to discover a sound binding hypothesis for a set of arylmethylaminephenyl derivatives. Taken together, both machine learning and structure-based approaches helped to study the inhibition of ABC-transporters by small molecules. Many questions remain open regarding the polyspecificity of many compounds for BCRP and P-glycoprotein, and especially around the actual three-dimensional structure of BCRP, which remains elusive even after the first cryo-electron microscopy map was published.