查看文章

Background

Neuroimaging abnormalities are critical components of psychopathology, and the study of shared genetic architecture between imaging and psychiatric phenotypes can yield important clues about etiopathophysiology. However, currently available imaging genomic datasets are complex, with thousands of image-derived phenotypes (IDPs) available, requiring computationally intensive data reduction techniques.

Methods

Employing UMAP (Uniform Manifold Approximation and Projection for Dimension Reduction) and DBScan (Density-Based Spatial Clustering of Applications with Noise), we distilled 1786 IDPs with significant heritability from the UK Biobank into coherent clusters. Subsequent genomic PCA enabled the generation of GWAS summary statistics for each cluster, bridging the gap between latent neuroimaging dimensions and their genetic underpinnings.

Results

We identified 15 IDP clusters …