Eligible Student Poster 49th Lorne Conference on Protein Structure and Function 2024

GPCRtm: a machine learning-based method for predicting the impact of mutations on G protein-coupled receptor stability (#304)

Joao Velloso 1 2 3 , Alex de Sa 1 2 3 4 , Douglas Pires 1 2 3 5 , David Ascher 1 2 3 4
  1. School of Chemistry and Molecular Biosciences, University of Queensland, Brisbane, Queensland, Australia
  2. Systems and Computational Biology, Bio21 Institute, University of Melbourne, Melbourne, Victoria, Australia
  3. Computational Biology and Clinical Informatics, Baker Heart and Diabetes Institute, Melbourne, Victoria, Australia
  4. Baker Department of Cardiometabolic Health, University of Melbourne, Melbourne, Victoria, Australia
  5. School of Computing and Information Systems, University of Melbourne, Melbourne, Victoria, Australia

G protein-coupled receptors (GPCRs) are key targets in drug discovery, with 30-40% of approved drugs targeting them. The structural determination of GPCRs, however, is challenging, primarily due to the inherent instability and intrinsic plasticity of GPCRs. GPCR engineering is one of the ways to handle this hurdle. An important example of GPCR engineering is the use of point mutations to minimise conformational heterogeneity and enhance crystal contacts and stability. Despite the critical role of mutations in stabilising GPCR structures, a lack of robust and accurate (experimental or computational) methodologies has still persisted, hindering advancements in GPCR research.

Given the cost, length and human dependence of experimental approaches, this study focuses on the development of computational methods to help guide rational GPCR engineering. We developed GPCRtm, a novel machine learning-based method to assess the impacts of mutations on GPCR stability. Through the collection of data on mutations in GPCRs and their known influence on protein stability (ΔTm), a supervised machine learning regressor was developed. This regressor utilises a mutation modelling approach, incorporating graph-based signatures and auxiliary features to characterise GPCR properties comprehensively. The resulting machine learning model is capable of predicting and ranking the effects of mutations on GPCR structure stability.

The results demonstrate that GPCRtm outperforms alternative methods while assessing mutation effects based on protein stability. It exhibits efficacy across various types of GPCRs, showcasing its versatility. The proposed machine learning model achieved correlation coefficients of 0.80 and 0.46 on 10-fold cross-validation over the training set and blind test set, respectively. Notably, when analysing the best features to describe the impact of mutations on stability, the graph-based signatures were considered as crucial features for predicting destabilising mutations, indicating their effectiveness in describing environmental changes caused by mutations.

GPCRtm advances our understanding of GPCR structures and also provides practical utility in drug discovery. By accurately ranking mutations and reducing the number of experimental assays needed, GPCRtm streamlines the elucidation of GPCR structures. It offers a robust, reliable, and easy-to-use web-based platform at https://biosig.lab.uq.edu.au/gpcr_tm/.