PhD Studentship: Developing a variant effect predictor for multiple missense variants in the same protein molecule to aid clinical interpretation

Context:

Genome sequencing is becoming increasingly widespread in both research and diagnostics, and over a million people have already had their genomes sequenced. However, interpreting the effect of genetic variants is challenging and frequently delays diagnosis in patients with rare genetic diseases. The number of variants of uncertain significance (VUS) in clinical databases is growing exponentially, and >70% of them are missense variants, i.e. where one amino acid in a protein sequence is replaced with another. Numerous computational variant effect predictors (VEPs) have been developed to predict the effect of missense variants and are increasingly being used in genome sequencing analysis and diagnostic interpretation. However, despite every human having ~10,000 missense variants in their genome, and ~1000 genes on average with more than one missense, genetic variant interpretation still largely considers these variants individually. Although some computational models can predict the effects of multiple amino acid changes, these have mainly been developed for protein-engineering purposes.

Key question:

The aim of this PhD project is to develop and validate a new VEP designed to predict the effect of multiple amino acid changes in the same protein to aid clinical interpretation.

Interdisciplinarity:

This is a highly interdisciplinary project that involves in silico 3D protein structure modelling and machine learning methods, working with genetic (e.g. clinical, population), biochemical (e.g. AlphaFold) and functional (e.g. MAVEs) datasets. We have therefore assembled a team of supervisors with experience in genetics, bioinformatics and protein structure, who provide complementary expertise:

- Exeter (Wright, Leslie): applying and assessing VEPs for clinical diagnosis and gene-disease discovery in rare disease

- Edinburgh (Marsh): comparing and validating VEPs against different datasets

- Imperial (David): designing and developing VEPs using 3D protein structure

Methodology:

1. Review and benchmark existing approaches for predicting multiple missense variants, including evolutionary, protein language and 3D structure-based models.
2. Develop existing methodology to predict the effect of one missense variant in the presence of another:a. Within the same structured protein domain; b. Within the same protein but in different domains; c. Within the same protein but including unstructured/interdomain regions; d. In different proteins within the same complex. Predictions will use information about 3D proximity, interactions between the amino acids, substrate binding, post-translational modifications, contribution to domain stability, and standard metrics of conservation, constraint, etc.
3. Test whether predictions capture interactions beyond the individual variant effects using existing data, e.g. MAVEs with double mutants, known examples of pathogenic alleles in cis, genome sequencing data in population cohorts (with statistical or read-based haplotyping), etc.
4. Possible extension: further development of the VEP to include prediction of three or more missense variants in the same molecule. Expected outcomes: The main outcome will be a new “multiVEP” for multiple missense variants in the same molecule that is validated against experimental and clinical data and available for use by the community.

Funded by: EPSRC Engineering and Physical Sciences Research Council