Author: Wentao Li, MS (2025)
Primary advisor: Xiaobo Zhou, PhD
Committee members: Jia Wu, PhD and Pora Kim, PhD
PhD thesis, McWilliams School of Biomedical Informatics at UTHealth Houston.
ABSTRACT
Integrating genetic data with medical imaging offers powerful opportunities to uncover biological mechanisms and clinical insights, yet linking genetic variation to diverse physical and molecular changes remains challenging. This dissertation introduces computational approaches that bridge this gap across neuroimaging and spatial omics. The first study develops a mixed-effects framework to identify genetic signatures of neuroimaging phenotypes in bipolar disorder. By modelling subject-specific brain region interactions, the method improves detection of significant variants in large-scale datasets such as the ABCD study and UK Biobank, with downstream analyses confirming biological relevance. The second study presents a cross-modal fusion strategy for spatial transcriptomics, integrating histology images with gene expression. Leveraging foundation models in different modalities with distillation and contrastive learning, this framework enhances imaging representations with gene-enriched information. The third study extends multiplex spatial proteomics by integrating proteomic data with scRNA using FuseMax to create pseudo–spatial transcriptomics, applying CellNEST to construct cell–cell interaction networks, and using these networks with limited protein panels to expand the limited proteomic coverage for better downstream performance. Collectively, these contributions advance integrative methodologies that connect genetics, imaging, and spatial omics, enabling deeper insights into disease mechanisms and the discovery of novel biomarkers.