A CNN Approach to Polygenic Risk Prediction of Kidney Stone Formation

TL;DR

This paper presents a CNN-based model that improves genetic risk prediction for kidney stone formation by modeling complex genetic interactions, outperforming traditional machine learning methods.

Contribution

It introduces a novel CNN approach to enhance polygenic risk scores for kidney stones, capturing non-linear genetic interactions from GWAS data.

Findings

CNN outperforms logistic regression, random forest, and SVM in accuracy and ROC-AUC.

Achieved 62% validation accuracy and 0.68 ROC-AUC.

Demonstrates potential of deep learning in genomics-based disease prediction.

Abstract

Kidney stones are a common and debilitating health issue, and genetic factors play a crucial role in determining susceptibility. While Genome-Wide Association Studies (GWAS) have identified numerous single nucleotide polymorphisms (SNPs) linked to kidney stone risk, translating these findings into effective clinical tools remains a challenge. In this study, we explore the potential of deep learning techniques, particularly Convolutional Neural Networks (CNNs), to enhance Polygenic Risk Score (PRS) models for predicting kidney stone susceptibility. Using a curated dataset of kidney stone-associated SNPs from a recent GWAS, we apply CNNs to model non-linear genetic interactions and improve prediction accuracy. Our approach includes SNP selection, genotype filtering, and model training using a dataset of 560 individuals, divided into training and testing subsets. We compare our CNN-based model with traditional machine learning models, including logistic regression, random forest, and support vector machines, demonstrating that the CNN outperforms these models in terms of classification accuracy and ROC-AUC. The proposed model achieved a validation accuracy of 62%, with an ROC-AUC of 0.68, suggesting its potential for improving genetic-based risk prediction for kidney stones. This study contributes to the growing field of genomics-driven precision medicine and highlights the promise of deep learning in enhancing PRS models for complex diseases.