Autoencoder-based phenotyping of ophthalmic images highlights genetic loci influencing retinal morphology and provides epidemiologically informative biomarkers.

Genome-wide association studies (GWAS) have been remarkably successful in identifying associations between genetic variation and imaging-derived phenotypes. To date, the main focus of these analyses has been established, clinically-used imaging features. Here, we sought to investigate if deep learning approaches can help detect more nuanced patterns of image variability. To this end, we used an autoencoder to represent retinal optical coherence tomography (OCT) images from 31,135 UK Biobank participants. For each study subject, we obtained a 64-dimensional vector representing features of retinal structure. GWAS of these autoencoder-derived imaging parameters identified 118 statistically significant loci; 17 of these associations also reached genome-wide significance in a replication analysis that included 10,409 UK Biobank volunteers. These loci encompassed variants previously linked with retinal thickness measurements, ophthalmic disorders and/or neurodegenerative conditions (including dementia). Notably, the generated retinal phenotypes were found to contribute to predictive models for glaucoma and cardiovascular disorders. Overall, we demonstrate that self-supervised phenotyping of OCT images enhances the discoverability of genetic factors influencing retinal morphology and provides epidemiologically informative biomarkers.