Ocular Microbiology and Immunology Group
Back to OMIG Main Page

Previous | 2024 Agenda and Abstracts | Next

2024 OMIG Abstract

Prediction of Subepithelial Infiltrates (SEI) following Adenovirus D8 Keratoconjunctivitis via Machine Learning

Behrouz Rahimi, Kenji Nakamichi, Russell N Van Gelder

Department of Ophthalmology and Karalis Johnson Retina Center,
University of Washington School of Medicine, Seattle, WA

Purpose: To validate a previous machine learning-based model for development of SEI on an additional group of samples.

Methods: Eleven previously unanalyzed conjunctival swabs from the treatment arm of the BayNovation auriclosene clinical trial that were positive for adenovirus D8 were sequenced by nanopore shotgun metagenomics and full sequences reconstructed. Sequences were analyzed using a previously trained random forest machine learning model (Nakamichi et al., Ophthalmology Science 2022) and predictions for SEI development compared with clinical outcomes at day 18.

Results: Adequate DNA for reconstruction of full viral genome was obtained in 11 samples. Analysis of sequence from these samples using the machine learning algorithm previously developed yielded the correct outcome in 9 of 11 samples, with two samples being false negative. When combined with 16 previously sequenced samples, the machine learning model correctly predicted outcome in previously unsequenced samples in 25 of 27 cases (accuracy 0.92, sensitivity 0.90, specificity 1.0, positive predictive value 100%, negative predictive value 75%).

Conclusions: Analysis of this additional cohort confirms that in this population, sequence variants of adenovirus D8 are largely responsible for development of SEI. Further validation in fully independent populations is warranted.


Disclosure:

N