CALM Aligns Brain Imaging and Genetics for Biomarker Discovery.

This research introduces CALM, a novel framework designed to uncover meaningful connections between brain structure and genetic influences, which are crucial for understanding neuropsychiatric conditions. A significant challenge in this field is that most large-scale datasets are unimodal, meaning they contain either neuroimaging data or genetic information, but rarely both for the same individuals. CALM addresses this by learning interpretable associations between brain regions of interest (ROIs) and genetic pathways even when the data comes from completely disjoint populations. The core of CALM involves aligning the two distinct modalities into a common latent space. This is achieved through linear projections that simultaneously ensure that the class-conditional latent distributions match and that different groups (e.g., healthy vs. diseased) remain separable. These projections are designed to be interpretable, directly revealing the pathway-ROI associations. When tested on unimodal imaging and genetics datasets, CALM demonstrated superior performance compared to existing state-of-the-art methods and baseline approaches, even generalizing effectively to previously unseen paired datasets. Experiments on autism spectrum disorder (ASD) using CALM identified immune and metabolic pathways linked to specific cortical regions, findings consistent with established scientific literature. This capability opens new avenues for leveraging vast unimodal data repositories to study complex cross-modal interactions in brain disorders.