Multi-modal machine learning of large-scale genetic, biologic, and brain imaging data for identification of therapeutic targets and early prediction b

Lead Research Organisation: Imperial College London
Department Name: Dept of Computing

Abstract

The aim of this project is to develop multi-modal models across data from high-throughput technologies such as genomics, metabolomics, proteomics, and radiomics, to identify genetic variations linked with cellular biochemical pathways leading to changes in brain structures and neurodegeneration. While several disparate data sources and population-based studies have been available in this space, drawing conclusive inferences across these datasets has been challenging for biostatisticians due to the large-scale and heterogeneous nature of these data. To date, AI/ML approaches have been severely underutilised. Biostatistical identification of single genetic hits without simultaneous modelling of the large array of metabolomics data that are markers of the linked biochemical pathways, yields genetic variants that are commonly in non-functional regions of the genome, have weak associations with the disease and do not provide useful insights. Likewise, identification of single metabolomic hits in isolation of the upstream genetic and downstream brain imaging features do not provide meaningful insights into the pathogenesis of dementia. We propose to combine the large array of genomics and metabolomics data with data on the intensity, shape and textural features of different brain structures derived from structural and functional MRI associated with neurodegeneration and dementia. Data for implementation of the project is available from the Chariot-Pro study (Number of participants=1200), the Alzheimer's Disease Neuroimaging Initiative (ADNI, N=1800) and UK Biobank(N\>20,000). All participants in these ongoing studies have contributed genomic and structural-MRI data, have been followed up for at least 5 years with repeated cognitive assessments and any diagnosis of dementia recorded. Functional-MRI and PET scans with multiple radioligands are additionally available in Chariot-Pro study and metabolomic and proteomic assays are underway. Additional data sources can be added to the project as they become available.

Publications

10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/S023283/1 01/04/2019 30/09/2027
2447401 Studentship EP/S023283/1 05/10/2020 04/10/2024 Xavier Cadet