Rationale
Physical inactivity is a leading modifiable risk factor for multiple diseases, yet the molecular routes by which exercise confers protection remain incompletely mapped. To address this gap, our research in recent years has focused on integrative multi-omics data analysis from exercise training experiments, yielding multiple novel insights. The UK Biobank (UKB) uniquely couples genotypes, multi-omics, high-resolution imaging, accelerometer-derived activity profiles and clinical follow-up for >500k participants, providing an unparalleled platform to dissect these mechanisms with both causal inference and multimodal AI.
Research questions
Which molecular signatures induced by exercise causally modify cardiometabolic and mental-health outcomes? Are exercise response maps enriched for disease risk factors?
Can a multimodal model integrating genome, imaging, accelerometer and clinical records identify polygenic signatures that involve exercise response transducers?
Can we utilize Mendelian Randomization (MR) analysis of pQTL data to learn protein-protein causal networks that explain cross-tissue communication patterns observed in preclinical exercise datasets?
Objectives
Deploy (multivariable) MR (single- or two-sample whenever appropriate) with putative cis pQTL instruments of exercise response tissue-specific proteins to quantify their causal effects on chronic diseases.
Develop a multimodal representation learning framework that disentangles modality specific unwanted variability. Train new models on MRI embeddings, DXA embeddings, metabolomics, and week-long accelerometry. Perform GWAS on selected representations and benchmark (e.g., via colocalization) on incident cardiometabolic events and VO!-max proxies.
Develop algorithms for causal network discovery among proteins and apply them on selected exercise-associated signatures. Develop strategies to test if discovered pathways align with exercise-induced cross-tissue communication patterns.
