Autoencoder-based phenotyping of ophthalmic images highlights genetic loci influencing retinal morphology and provides informative biomarkers
Last updated:
Author(s):
Panagiotis I Sergouniotis, Adam Diakite, Kumar Gaurav, Naomi Allen, Tariq Aslam, Denize Atan, Sarah Barman, Jenny Barrett, Paul Bishop, Graeme Black, Tasanee Braithwaite, Roxana Carare, Usha Chakravarthy, Michelle Chan, Sharon Chua, Alexander Day, Parul Desai, Bal Dhillon, Andrew Dick, Alexander Doney, Cathy Egan, Sarah Ennis, Paul Foster, Marcus Fruttiger, John Gallacher, David Garway-Heath, Jane Gibson, Jeremy Guggenheim, Chris Hammond, Alison Hardcastle, Simon Harding, Ruth Hogg, Pirro Hysi, Pearse Keane, Peng Tee Khaw, Anthony Khawaja, Gerassimos Lascaratos, Thomas Littlejohns, Andrew Lotery, Robert Luben, Phil Luthert, Tom Macgillivray, Sarah Mackie, Savita Madhusudhan, Bernadette Mcguinness, Gareth Mckay, Martin Mckibbin, Tony Moore, James Morgan, Eoin O'Sullivan, Richard Oram, Chris Owen, Praveen Patel, Euan Paterson, Tunde Peto, Axel Petzold, Nikolas Pontikos, Jugnoo Rahi, Alicja Rudnicka, Naveed Sattar, Jay Self, Panagiotis Sergouniotis, Sobha Sivaprasad, David Steel, Irene Stratton, Nicholas Strouthidis, Cathie Sudlow, Zihan Sun, Robyn Tapp, Dhanes Thomas, Emanuele Trucco, Adnan Tufail, Ananth Viswanathan, Veronique Vitart, Mike Weedon, Cathy Williams, Katie Williams, Jayne Woodside, Max Yates, Jennifer Yip, Yalin Zheng, Ewan Birney, Tomas Fitzgerald
MOTIVATION: Genome-wide association studies (GWAS) have been remarkably successful in identifying associations between genetic variants and imaging-derived phenotypes. To date, the main focus of these analyses has been on established, clinically-used imaging features. We sought to investigate if deep learning approaches can detect more nuanced patterns of image variability.
RESULTS: We used an autoencoder to represent retinal optical coherence tomography (OCT) images from 31 135 UK Biobank participants. For each subject, we obtained a 64-dimensional vector representing features of retinal structure. GWAS of these autoencoder-derived imaging parameters identified 118 statistically significant loci; 41 of these associations were also significant in a replication study. These loci encompassed variants previously linked with retinal thickness measurements, ophthalmic disorders, and/or neurodegenerative conditions. 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.
AVAILABILITY AND IMPLEMENTATION: Code and data links available at https://github.com/tf2/autoencoder-oct.
Optical coherence tomography imaging (OCT) rapidly produces 3 dimensional images of the macula (the sensitive part of the retina used for central vision). Abnormalities of…
Institution:
Moorfields Eye Hospital NHS Foundation Trust, Great Britain
Copy number variation (CNV) is an important source of genetic differences between individuals and can give raise to observable traits. Certain CNVs are known to…
Institution:
European Bioinformatics Institute (EBI), Great Britain
Rare eye disorders are a major cause of visual impairment and blindness in children and young adults. Over 1000 such disorders have been described (most…
