About the Project
G protein coupled receptors (GPCRs) are involved in a large number of the body’s signalling pathways. Evidence is accumulating that demonstrates that drugs targeting GPCRs activate multiple separate desirable and undesirable signalling pathways that mediate their pharmacological effects, reducing the efficacy of the therapy and increasing undesirable side effects. Our hypothesis is that it is necessary to understand the molecular basis of drug action in GPCRs in order to efficiently and rationally design better molecules with biased signalling profiles that promote favourable outcomes and do not trigger negative ones. We propose that a combination of unique experimental data and modern machine learning approaches provide a high potential route to developing this understanding.
We will develop state-of-the art AI and machine learning techniques to elucidate the mechanisms of drug action in GPCRs, using unique data from Veprintsev’s high-throughput alanine scanning mutagenesis technique to study the role that individual residues in GPCRs play in signalling pathways in response to different ligands (Sun et al 2015). This has been used to discover a universal allosteric mechanism for a specific pathway (Flock et al, 2015) and recent work has identified the molecular basis of biased signaling by a GPCR vasopressin receptor by profiling a library of 400 alanine mutants and generating 12000 concentration-response activation curves, measuring the activation of all classes of G proteins and arrestins. We showed how individual mutations induce bias towards specific signaling pathways, and proposed specific states that explain the origin of this.
Elementary machine learning methods have been used to good effect in these studies, but are not sufficiently expressive to take advantage of the quantity of data available, or to incorporate existing knowledge of GPCR structure and function. We will develop novel machine learning approaches to model the complex relationships between the many interdependent variables, and incorporate our existing knowledge of receptor structure and function, and to predict the contribution of individual residues to signalling. We will incorporate structural information to enable the model to generalise and predict the contribution of residues to signalling in other receptors. Data from high-throughput screens of small molecular compound libraries will allow us to predict the action of a large range of drugs across a variety of GPCR classes.
Finally, we will investigate “inversion” of these models to predict a ligand that will give rise to a desired signalling profile. For this, modern approaches to generative modelling are likely to play a key role.
Applicants should have a strong background in mathematics and computation, and ideally a background in machine learning and an interest in biology. They should have a commitment to research in machine learning and hold or realistically expect to obtain at least an Upper Second Class Honours Degree in computer science, mathematics, physics, or a similar numerate discipline with a significant programming component.
2. Tilman Flock, Charles NJ Ravarani, Dawei Sun, Aiveliagaram J Venkatakrishnan, Melis Kayikci, Christopher G Tate, Dmitry B Veprintsev, M Madan Babu. Universal allosteric mechanism for Gα activation by GPCRs. Nature 524 (7564), 173 (2015)
Based on your current searches we recommend the following search filters.
Based on your current search criteria we thought you might be interested in these.