Preview

PDF, English Download (8MB) | Terms of use

Abstract

Genetic variations can have positive, negative, or neutral impacts on protein interactions, thus making it essential to understand them to obtain a mechanistic picture of biological functions and diseases. In this thesis, I studied how genetic changes affect the functions of the largest, most diverse family of cell-surface molecules involved in signal transduction: G-protein coupled receptors (GPCRs). GPCRs comprise over 2% of genes in the human genome and are the leading pharmaceutical drug target.

Analysis of one of the most comprehensive datasets quantifying GPCR/G-protein binding affinities revealed that GPCR couplings and sequence similarity are uncorrelated and that there were no clear, simple sequence changes responsible for determining which G-protein binds to a particular GPCR. While GPCRs within the same group can couple to different G-proteins, GPCRs of different groups can still couple to the same G-proteins. We used this new dataset and various protein bioinformatics tools to identify broad sequence features that are associated with specific G-protein binding events. Several of these were at or near the known GPCR/G-protein interface, but many others were not, suggesting a complex relationship between sequence and specificity.

We then applied an interpretable machine learning algorithm on the sequence- and structure-based GPCR features to develop a system and associated webserver (PRECOG) to predict and visualize GPCR/G-protein interactions. We leveraged the machine learning-guided framework to predict uncharacterized GPCRs and successfully developed the first GNA12-coupled designer receptor. Application of this framework to recently available binding data revealed the determinants of β-arrestin specificity in GPCRs.

Collectively, this machine learning-guided framework can be extended to other binding data to uncover sites and sequence regions that are physically or allosterically involved in determining subtype specificity. This will not only improve our understanding of protein interactions but also help us devise better chemogenetic tools and take smarter therapeutic decisions in the context of human health.