Predicting Diabetic Retinopathy Using a Machine Learning Approach Informed by Whole-Exome Sequencing Studies

Area of Science:

• Genomics and Precision Medicine
• Ophthalmology
• Computational Biology

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss in diabetic patients.
• Early detection and risk prediction are crucial for effective management and prevention of DR-related complications.
• Current risk prediction models often lack comprehensive genetic profiling.

Purpose of the Study:

• To develop and validate a novel risk-prediction model for diabetic retinopathy (DR).
• To utilize whole-exome sequencing (WES) and machine learning (ML) for enhanced DR risk assessment.
• To identify specific genetic variants associated with DR susceptibility.

Main Methods:

• Whole-exome sequencing (WES) was performed on a DR pedigree to identify single nucleotide polymorphisms (SNPs) and mutations.
• A machine learning (ML) prediction model was built and validated using genetic and demographic data from 420 type 2 diabetic patients.
• Shapley Additive explanation analysis was employed to assess feature contributions, comparing models with and without identified SNPs.

Main Results:

• Seven SNPs/mutations were found to be associated with DR, including variants in TRIM7, LRBA, PRMT10, C9orf152, CLDN25, SH3GLB2, and FANCC.
• The prediction model incorporating rs146694895 and rs201407189 demonstrated superior performance.
• The enhanced model achieved an accuracy of 80.2%, sensitivity of 83.3%, specificity of 76.7%, and an AUC of 80.0%.

Conclusions:

• Novel single nucleotide polymorphism (SNP) sites associated with diabetic retinopathy (DR) were identified.
• The inclusion of specific SNPs (rs146694895 and rs201407189) significantly improved the predictive power of the machine learning model.
• This WES-based ML approach offers a promising tool for personalized DR risk prediction.