Enhancing Glaucoma Risk Prediction through Advanced Genomics and Machine Learning - Project Summary/Abstract Glaucoma is a multifactorial disease affecting millions worldwide, leading to substantial visual impairment and blindness. The lack of reliable tools to predict disease risk and progression has resulted in preventable vision loss and inefficient use of medical resources. In recent years, polygenic risk scores (PRS), which quantify the cumulative risk of disease based on multiple genetic variants, offer a promising approach for personalized risk prediction. However, the performance of these scores remains moderate at best limiting their clinical use. This project aims to improve PRS performance using a multipronged approach combining innovative genomics and computational strategies, thereby improving individualized glaucoma risk assessment and prognostication. We will achieve this goal through the following the following 3 specific aims: 1) Improve PRS Performance through Enhanced EHR-Based Disease Phenotyping: leveraging cutting-edge Natural Language Processing (NLP) techniques and data from 68000+ patients across two clinical centers, we will train models to extract refined glaucoma phenotypes from unstructured EHR data and demonstrate improved performance compared to traditional ICD-based phenotyping. Results from genome-wide association study (GWAS) of our NLP phenotypes from over 1 million participants across two biobanks will be used to develop improved PRS for glaucoma. 2) Utilize ML-Derived Structural Phenotypes to Enhance PRS Performance: by conducting GWAS on our previously developed machine learning-derived Optical Coherence Tomography (OCT) glaucoma phenotypes from 3 biobanks including 53000+ subjects, we will pinpoint novel genetic variants and integrate these variants with previously known variants to create PRS with improved performance. 3) Improve PRS performance by Constructing Subtype and Pathway-Specific PRS Models: with genomic data from 312,944 All of Us participants, we will develop specialized PRS models for open-angle glaucoma subtypes as well as pathway-specific scores based on relevant biologic pathways. These models, informed by previous GWAS results, will enhance our understanding of glaucoma's complex genetic risk, and better predict disease severity, trajectory, and treatment response. The culmination of these innovative methodologies seeks to transform glaucoma care, promoting precision in early detection and tailored management. This interdisciplinary approach, merging computational expertise with robust genomic insights, holds promise to set new standards in ophthalmologic precision medicine.
