Development and Application of Virtual Screening Methods for G Protein-Coupled Receptors

Abstract

G protein-coupled receptors (GPCR) constitute one of the largest family of transmembrane proteins that have been implicated in a multitude of diseases, including cancer and diabetes, and have been an important target in drug deve lopment. While experiment-based high-throughput screening for the unearthing of novel chemical compounds remains the de facto standard for drug discovery, virtual screening has been gaining acceptance as an important complementary method due to its high speed and low cost, which instead employs computers.

This dissertation is aimed at the development of virtual screening algorithms as applied to GPCR’s, in addition to the construction of GPCR-related databases (GPCR-EXP, GLASS). MAGELLAN is a ligand-based virtual screening algorithm that makes inferences about what a GPCR would potentially bind based on sequence- and structure-based alignments. Building on top of this work, a sequential virtual screening pipeline combining MAGELLAN with AutoDock Vina was constructed for the discovery of novel, bifunctional opioids with mu opioid receptor (MOR) agonist and delta opioid receptor (DOR) antagonist activity.

In the process of developing the virtual screening algorithms, two GPCR-related databases were constructed to provide necessary data for the study. GPCR-EXP is a database of experimentally-validated and predicted GPCR structures. Important features include semi-manual curation of data, weekly updates, a user-friendly web interface, and high-resolution structure models with GPCR-I-TASSER, which many of the other GPCR-related databases lack. Additionally, GLASS database was developed in response to the absence of databases dedicated to GPCR experimental data. As a result, pharmacological data was pooled and integrated into a single source, resulting in over 500,000 unique GPCR-ligand associations; this made it the most comprehensive database of its kind thus far, providing the community with an accessible web interface, freely-available data, and ligands ready for docking.

MAGELLAN utilized pharmacological data from GLASS to infer from the ligands of sequence- and structure-based homologues what a target GPCR would bind. It was tested on two public virtual screening databases (DUD-E and GPCR-Bench) and achieved an average EF of 9.75 and 13.70, respectively, which compared favorably with AutoDock Vina (1.48/3.16), DOCK 6 (2.12/3.47), and PoLi (2.2). Lastly, case studies with the mu opioid and motilin receptors demonstrated its applicability to virtual screening in general, as well as GPCR de-orphanization. Subsequently, MAGELLAN was combined with AutoDock Vina into a novel, sequential virtual screen pipeline against both MOR and DOR to compensate for the weaknesses of each algorithm. Retrospective virtual screens against both MAGELLAN and AutoDock Vina were established for both receptors, and both methods were reported to have over-random discrimination between actives and decoys using the GPCR-Bench dataset.

In conclusion, structure (GPCR-EXP) and pharmacological data (GLASS) databases were constructed to provide users with a comprehensive source of GPCR data. Moreover, GLASS made it possible for MAGELLAN to be developed, providing it a rich source of experimental data. In return, this resulted in greater performance than competing algorithms. Lastly, a prospective sequential virtual screening pipeline was established for the discovery of novel bifunctional opioids, in which the models for both methods were validated to perform well. In future studies, cAMP and β-arrestin assays will be run on a subset of compounds from a prospective virtual screen in the hopes of discovering a novel opioid with reduced tolerance and withdrawal.