Accuracy of genome-enabled polygenic risk score prediction of cruciate ligament rupture risk in the Labrador Retriever.

INTRODUCTION

Canine cruciate ligament rupture (CR) is a common, complex, polygenic, orthopaedic disease in dogs that results in serious financial burden and patient morbidity even in the face of surgical correction. The goal of this study was to evaluate the clinical utility of CR polygenic risk score (PRS) prediction models using genome-wide SNP data from a large reference population of Labrador Retriever dogs.

METHODS

Using 10-fold cross-validation and an independent validation population, we assessed Bayesian and machine learning models with and without covariates using both genome-wide SNPs as well as genic SNPs. Models were tuned by optimizing numbers of CR risk SNPs selected by genome-wide association and adjusting posterior probability thresholds to maximize prediction accuracy.

RESULTS

Models that included clinical covariates (sex, neuter status, age, weight, withers height, as well as the first 10 principal components from the genetic relationship matrix) universally yielded higher accuracy up to 88.5% compared to 77% without covariates. Prediction accuracy for some models was reduced when only genic SNPs were used suggesting SNPs in non-coding regions could influence the CR disease risk.

DISCUSSION

Our results confirm that PRS models provide sufficient predictive accuracy for clinical application in veterinary medicine and offer a viable, early-life screening tool for personalized care and selective breeding to reduce CR incidence in high-risk breeds. Our results further confirm that CR is a complex polygenic disease in which genome-wide risk SNPs influence disease pathogenesis.