Towards Personalized Medicine: Computational Approaches to Support Drug Design and Clinical Decision Making
Charlotta Schärfe · Universitätsbibliothek Tübingen · 2019
The future looks bright for a clinical practice that tailors the therapy with the best efficacy and highest safety to a patient. Substantial amounts of funding have resulted in technological advances regarding patient-centered data acquisition --- particularly genetic data. Yet, the challenge of translating this data into clinical practice remains open. To support drug target characterization, we developed a global maximum entropy-based method that predicts protein-protein complexes including the three-dimensional structure of their interface from sequence data. To further speed up the drug development process, we present methods to reposition drugs with established safety profiles to new indications leveraging paths in cellular interaction networks. We validated both methods on known data, demonstrating their ability to recapitulate known protein complexes and drug-indication pairs, respectively. After studying the extent and characteristics of genetic variation with a predicted impact on protein function across 60,607 individuals, we showed that most patients carry variants in drug-related genes. However, for the majority of variants, their impact on drug efficacy remains unknown. To inform personalized treatment decisions, it is thus crucial to first collate knowledge from open data sources about known variant effects and to then close the knowledge gaps for variants whose effect on drug binding is still not characterized. Here, we built an automated annotation pipeline for patient-specific variants whose value we illustrate for a set of patients with hepatocellular carcinoma. We further developed a molecular modeling protocol to predict changes in binding affinity in proteins with genetic variants which we evaluated for several clinically relevant protein kinases. Overall, we expect that each presented method has the potential to advance personalized medicine by closing knowledge gaps about protein interactions and genetic variation in drug-related genes. To reach clinical applicability, challenges with data availability need to be overcome and prediction performance should be validated experimentally.