A Deep Learning Approach to Polygenic Risk Prediction of Kidney Stone Formation

Kidney stone disease affects millions of people worldwide, with both environmental and genetic factors contributing to individual susceptibility. While Genome-Wide Association Studies (GWAS) have successfully identified numerous single-nucleotide polymorphisms (SNPs) associated with kidney stone risk, translating these findings into accurate and clinically useful prediction tools remains a significant challenge. Polygenic Risk Scores (PRS) provide a framework for quantifying an individual’s genetic predisposition, but traditional PRS approaches often fail to account for complex interactions among variants and the non-linear effects present in genomic data. In this study, we investigate the potential of deep learning techniques, specifically Convolutional Neural Networks (CNNs), to improve PRS models for predicting kidney stone susceptibility. Our approach integrates careful SNP selection, genotype filtering, and CNN-based modeling to address challenges such as data imbalance, genomic noise, and high-dimensional feature spaces. We benchmark our CNN-based framework against conventional machine learning models, including logistic regression, random forests, and support vector machines, demonstrating superior performance in terms of classification accuracy and ROC-AUC. These findings underscore the promise of deep learning-enhanced PRS models to provide more accurate genetic risk predictions for kidney stones. The research highlights the potential of integrating advanced computational approaches with genomics to advance precision medicine, improve patient stratification, and guide preventive strategies for at-risk individuals.