The development of new drugs is a highly expensive and time-consuming endeavor, often plagued by failed clinical trials. To address these challenges, a wide array of computer-aided drug design (CADD) methods has been developed to guide and accelerate drug discovery. However, the predictive validity of these methods is often limited, resulting in their failure to accurately reproduce and predict experimental outcomes, thus raising questions about their overall value.
The advent of ultra-large make-on-demand libraries, containing millions of molecules, has introduced new value to virtual screening tools by enhancing the value of large-scale prioritizing and optimizing compounds for experimental testing. This thesis focuses on the implementation and development of tools designed to accelerate the virtual screening of billion-sized chemical spaces in the search for new inhibitors of P-glycoprotein (P-gp) - an ATP-powered transport protein involved in the cellular efflux of many drugs. P-gp's role in multidrug resistance (MDR) in cancer has made it a target for many drug discovery efforts since its discovery 50 years ago.
Given that many drug discovery campaigns targeting P-gp have resulted in large hydrophobic compounds that bind to the promiscuous substrate binding site of P-gp but fail to demonstrate clinical activity, this thesis explores the potential of virtually screening billions of compounds for their affinity towards potential binding sites on the nucleotide-binding domains (NBDs) of P-gp. Specifically, these virtual screens are accelerated using active learning (AL) and ramp-up AL to identify compounds with excellent docking scores towards five different sites. Notably, docking against one of these sites has yielded particularly promising lead-like molecules, with the top compound displaying a sub micromolar IC50 value in a Calcein-AM assay, making it a promising template for further hit-to-lead optimization.
In addition to introducing the ramp-up AL strategy, this thesis also presents SpaceGA - a novel screening approach that combines a genetic algorithm (GA) with the similarity search tool SpaceLight to perform docking screens in trillion-sized spaces. Space-GA thus represents a promising new tool for fully leveraging the growth of commercial libraries in the pursuit of more effective P-gp inhibitors.