MRC DiMeN Doctoral Training Partnership: Tensor-based machine learning for personalised medicine

This interdisciplinary project combines the disciplines of computer science, medicine and molecular biology to make technological advances in the emerging field of personalised medicine. The student will be supervised by early-career genomic medicine expert, Dr Dennis Wang, and early-career machine learning expert, Dr Haiping Lu. She/he will test the hypothesis that relationships between gene mutations can be used to predict drug responses in patients with neurodegenerative diseases.

A critical challenge in genetic and drug response data is high dimensionality but low sample size. Machine learning with tensor (multi-dimensional array) modelling will be investigated for significant methodology development. This will involve three stages of development: 1) Preprocess genomic and clinical data (from the 100K Genomes Project and UK Biobank) to enable machine learning – significant efforts are needed to convert raw data into a standard, high-quality form ready for machine learning;
2) Apply/adapt tensor-based co-/clustering on preprocessed data to discover higher-order relationships between genetic features and drug responses;
3) Analyse and interpret important genetic features for such relationships and validate them by predicting drug response results from the NIHR Sheffield Biomedical Research Centre (BRC) and AstraZeneca.

The student will also collaborate with other scientists at the BRC and AstraZeneca to define new patient phenotypes from the datasets, such as using machine learning to extract phenotypes from brain fMRI images. Expected achievements include a proof of principle for using higher-order relationships between genes and clinical phenotypes to predict drug response. By focusing on neurodegenerative diseases, the student will identify combinations of genetic mutations, as biomarkers to predict response to drugs (e.g. riluzole). The methodology developed in this project will be packaged into an open-source software accessible through Github.