June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Network-based machine learning for gene prioritization in primary open-angle glaucoma
Author Affiliations & Notes
  • Henry Corbett Cousins
    Biomedical Data Science, Stanford University School of Medicine, Stanford, California, United States
  • Russ B Altman
    Biomedical Data Science, Stanford University School of Medicine, Stanford, California, United States
  • Louis R Pasquale
    Ophthalmology, Icahn School of Medicine at Mount Sinai, New York, New York, United States
  • Footnotes
    Commercial Relationships   Henry Cousins None; Russ Altman None; Louis Pasquale Eyenovia, Skye Biosciences, Twenty Twenty, Character Bio, Code C (Consultant/Contractor)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2023, Vol.64, OD38. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Henry Corbett Cousins, Russ B Altman, Louis R Pasquale; Network-based machine learning for gene prioritization in primary open-angle glaucoma. Invest. Ophthalmol. Vis. Sci. 2023;64(8):OD38.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Primary open-angle glaucoma (POAG) has a strong, multifactorial genetic basis that remains incompletely understood. While hundreds of risk genes have been identified through associative and experimental studies, a considerable percentage of heritability remains unexplained, and prioritizing known risk genes for therapeutic efforts remains challenging. We hypothesized that network-based representation learning could facilitate prioritization of glaucoma risk genes based on protein-protein interactions (PPIs).

Methods : We obtained high-confidence, literature-derived PPIs from the STRING database (version 11.1) and 294 known POAG risk-associated genes supported by experimental or genome-wide association data from the DisGeNET database (version 7.0). We constructed an unweighted PPI network comprising 12,396 protein-coding genes as vertices and 324,152 PPIs as edges and performed low-dimensional feature learning using the node2vec algorithm. We then trained an L2-regularized logistic regression model on the resulting gene embeddings to classify labeled POAG risk genes using Monte Carlo cross-validation. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC). Pathway analysis on gene prioritization scores was performed using gene set enrichment analysis (GSEA).

Results : The model produced POAG priority scores for 12,396 human protein-coding genes using only the latent structure of the PPI network (Table). These scores correctly identified known POAG risk genes with an AUC of 0.745 (95% confidence interval (CI): 0.686-0.792). High-scoring genes had strong associations with inflammatory eye conditions in addition to POAG, including retinal degeneration (p < 0.001) and exophthalmos (p = 0.004). High-scoring genes demonstrated enrichment of several pathways with known involvement in POAG pathogenesis, including steroid hormone biosynthesis (p < 0.001), complement cascade (p < 0.001), and cytokine signaling (p < 0.001) (Figure).

Conclusions : Unsupervised network learning leveraging large-scale PPI datasets may provide a means of identifying and prioritizing candidate risk genes in POAG in the absence of disease-specific evidence regarding the gene itself.

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

 

Table. Ten highest-scoring genes based on model predictions

Table. Ten highest-scoring genes based on model predictions

 

Figure. Pathway enrichment among computationally predicted risk genes.

Figure. Pathway enrichment among computationally predicted risk genes.

×
×

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.

×