Investigating the genetics of deep learning derived neuro imaging phenotypes of brain disorders
O’Connell, Shane
O’Connell, Shane
Loading...
Repository DOI
Publication Date
2024-04-17
Type
Thesis
Downloads
Citation
Abstract
Brain disorders are collections of debilitating phenotypes that can affect cognition and general life quality via a myriad of symptoms, including mood swings, memory loss, altered thought processes, and psychosis. Despite their common area of action in the brain, few biomarkers have been characterised. Further, the causal relationship between neuroimaging measures and brain disorders remains largely un explored. Understanding the biological manifestations of these conditions could help to inform improved diagnostic, prognostic, and treatment systems. To this end, we identified neuroimaging biomarkers of Alzheimer’s disease using a convolutional neural network, and found 7 genome-wide significant loci asso ciated with the resultant quantity. These findings were consistent with previously observed genetic results of Alzheimer’s disease and further implicated impaired cellular homeostasis as a molecular association of Alzheimer’s disease-related neuroanatomical variation. We also trained an autoencoder on participant tabular neuroimaging data from the same dataset and highlighted a latent space node significantly asso ciated with Alzheimer’s participants, finding three genome-wide significant loci mapping to non-coding RNA transcripts associated with its value. Across both studies, we also demonstrate evidence of tissue specific expression in clinically relevant brain regions, such as the substantia nigra. We finally queried the causal relationship between neuroimaging measures and bipolar disorder using graph-based Mendelian randomization methods, finding that white matter microstructural phenotypes exert greater effects in a network context than gray matter structural phenotypes. Specifically, we find evidence of bidirectional causality between bipolar disorder and the area of the lateral orbitofrontal cortex and several components of the limbic system involved in emotional regulation. Taken together, our results provide novel avenues of enquiry into the derivation of neuroanatomical biomarkers and the investigation of causal dynamics between neuroimaging and brain disorders.
Funder
Publisher
NUI Galway