June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Genome wide discovery via feature space mapping of deep-learning derived clinical OCT phenotypes to the UK biobank
Author Affiliations & Notes
  • Nazlee Zebardast
    Massachusetts Eye and Ear Department of Ophthalmology, Boston, Massachusetts, United States
  • Saber Kazeminasab Hashemabad
    Massachusetts Eye and Ear Department of Ophthalmology, Boston, Massachusetts, United States
  • Yan Zhao
    Massachusetts Eye and Ear Department of Ophthalmology, Boston, Massachusetts, United States
  • Sayuri Sekimitsu
    Massachusetts Eye and Ear Department of Ophthalmology, Boston, Massachusetts, United States
  • Kanza Aziz
    Massachusetts Eye and Ear Department of Ophthalmology, Boston, Massachusetts, United States
  • Mohammad Eslami
    Massachusetts Eye and Ear Department of Ophthalmology, Boston, Massachusetts, United States
  • Mengyu Wang
    Massachusetts Eye and Ear Department of Ophthalmology, Boston, Massachusetts, United States
  • Tobias Elze
    Massachusetts Eye and Ear Department of Ophthalmology, Boston, Massachusetts, United States
  • Janey L Wiggs
    Massachusetts Eye and Ear Department of Ophthalmology, Boston, Massachusetts, United States
  • Footnotes
    Commercial Relationships   Nazlee Zebardast None; Saber Kazeminasab Hashemabad None; Yan Zhao None; Sayuri Sekimitsu None; Kanza Aziz None; Mohammad Eslami None; Mengyu Wang None; Tobias Elze None; Janey Wiggs None
  • Footnotes
    Support  NIH/NEI P30EY014104 ; K23EY032634
Investigative Ophthalmology & Visual Science June 2023, Vol.64, 980. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Nazlee Zebardast, Saber Kazeminasab Hashemabad, Yan Zhao, Sayuri Sekimitsu, Kanza Aziz, Mohammad Eslami, Mengyu Wang, Tobias Elze, Janey L Wiggs; Genome wide discovery via feature space mapping of deep-learning derived clinical OCT phenotypes to the UK biobank. Invest. Ophthalmol. Vis. Sci. 2023;64(8):980.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : Large population-based biobanks provide a rich source of genetic data but due to the nature of collected data, clinical relevance may be limited. Here we use a large clinical glaucoma dataset to discover macular OCT-based phenotypes using unsupervised deep learning (DL) and subsequently test these phenotypes in the UK Biobank to allow for genome wide discovery

Methods : Ganglion cell complex (GCC) thickness maps of 7966 high-quality macular optical coherence tomography (OCT) scans from 4926 glaucoma patients in Mass Eye Ear (MEE) dataset were used to train a DL model based on unsupervised representation learning, data augmentation and contrastive learning. After training, an encoder associated a 3-channel image into a 128-element feature vector. Dimensionality reduction with UMAP and Gaussian Mixture Model clustering was used to group semantically similar GCC patterns. Optimal number of clusters was determined using probabilistic model selection. The trained DL model on the MEE dataset was used to map 86115 GCC thickness maps of 50115 UKBB subjects into the feature space. We performed GWAS for each GMM-identified feature dimension using linear mixed models adjusted for age, sex, and 10 principal components of ancestry

Results : We identified 11 optimal macular GCC patterns (q0-10) in MEE dataset which were successfully mapped to UKBB. The distribution of GCC phenotypes in UKBB is shown in Fig1 except for q3 which was not found. GWAS was performed with 9329555 imputed genotypes after quality control. We found no evidence of unaccounted stratification (lambdaGC 1.000-1.0208). We identified several (range: 1-15) independently associated SNPs (P < 5e-08) for each phenotype (Fig2) and numerous genomic regions associated with at least one variant/phenotype related to glucocorticoid interaction (q5, nearest gene (NG): MDFIC), neurological development/disease (q5, NG: ATRNL1; q6, NG: SHISA9, MEIS3; q9, NG: ATP8A2; q10, NG: ZNF423), cholesterol synthesis (q6, NG: TM7SF2), immune system (q6, NG: IGLL5), and malignancy (q9, NG: ETV7, PCTP; q10, NG: PARD3B, ZNF423). The genomic regions were enriched for association with interferon-I activity (91 genes) based on MAGMA (FDR<0.05).

Conclusions : We demonstrate that OCT derived DL phenotypes discovered in a clinical cohort can be successfully transferred to an independent large biobank allowing for potential discovery of disease-relevant variants

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

 

 

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×