An Explainable Deep Learning Framework for Imaging Genetics: Deriving Brain-Genotype Scores From MRI to Link Genetic Variation, Brain Structure, and Cognition

Imaging genetics aims to understand how genetic variation influences brain structure and cognitive function. Traditional approaches often rely on imaging-derived phenotypes (IDPs), which reduce high-dimensional brain images to predefined summary measures and may miss subtle or spatially distributed genotype-related effects. We developed brain-genotype scores, continuous image-based representations of genetic variation learned directly from structural MRI. Using a multitask deep-learning framework trained on T1-weighted MRI from the UK Biobank, we predicted genotype dosage for 120 SNPs. The resulting genotype probability estimates were used as continuous brain-genotype scores. Unlike conventional IDPs, these scores are learned directly from raw images and capture distributed neuroanatomical patterns associated with genetic variants. Gradient-based saliency maps were used to localise neuroanatomical regions contributing to each score, providing interpretable links between genetic variation and brain anatomy. To evaluate their biological relevance, brain-genotype scores generated from an independent, held-out test set were used as neuroanatomical markers in association analyses with seven cognitive phenotypes, adjusted for population structure and technical covariates. A total of 147 out of 840 score-based associations survived FDR correction. In contrast, corresponding analyses using the original genotype data and traditional machine-learning-based scores trained on IDPs identified only two significant associations in total. These findings suggest that brain-genotype scores capture cognition-related neuroanatomical information beyond that available from genotype data alone, providing an interpretable, individual-level representation of genotype-related brain variation. By encoding both genetic and neuroanatomical information in SNP-specific brain-genotype scores, the framework may extend beyond cognition to other phenotypes, offering a new route to genotype-phenotype discovery in imaging genetics.