An Exploratory Study of Machine Learning-Based Open-Angle Glaucoma Detection Using Specific Autoantibodies.

Area of Science:

• Ophthalmology
• Immunology
• Bioinformatics

Background:

• Four autoantibodies (anti-ETNK1, anti-VMAC, anti-NEXN, anti-SUN1) were previously identified as associated with open-angle glaucoma (OAG).
• The diagnostic utility of these autoantibodies requires further evaluation.

Purpose of the Study:

• To assess the diagnostic performance of four OAG-associated autoantibodies using automated machine learning.
• To develop and validate a machine learning model for OAG diagnosis based on autoantibody levels.

Main Methods:

• Plasma samples from 119 OAG patients and 35 controls were analyzed.
• Automated machine learning (Python, scikit-learn, PyCaret) was employed, incorporating autoantibody levels, age, sex, and intra-ocular pressure.
• Model performance was evaluated using ROC-AUC, average precision, and F1 score, with probability calibration and explainability analyses.

Main Results:

• The random forest model achieved an out-of-fold ROC-AUC of 0.852, average precision of 0.950, and F1 score of 0.865.
• Isotonic mapping enhanced the agreement between predicted and empirical probabilities.
• Anti-VMAC autoantibody emerged as the most significant predictor in the model.

Conclusions:

• A machine learning model incorporating four autoantibodies shows potential for diagnosing OAG from blood samples.
• This approach may offer a novel, non-invasive method for OAG detection.