Immune system, inflammation and cardiovascular diseases: can we use autoimmune disease genetics to find clues?

Sammanfattning

Early and accurate prediction of genetic predisposition to cardiovascular disease is essential for monitoring and treating high-risk individuals. Previous studies on cardiovascular disease have shown that polygenic risk scores may aid in quantifying an individual’s inherited susceptibility to these diseases. While common approaches focus on using cardiovascular-associated variants in risk score calculation, we sought to develop these scores using autoimmune-related variants[DL1] . We developed a machine-learning-based workflow to identify independent and significantly associated variants with 10 cardiovascular outcomes in the Swedish EpiHealth cohort. This workflow aims to identify outcome-associated variants through feature selection, followed by a random-sample-based gradient-boosting model, polygenic risk score construction, and clinical covariate modelling. The models predicted moderate-to-strong discrimination between individuals with and without cardiovascular outcomes using a set of well-known risk factors, clinical covariates[DL2] , and polygenic risk scores in both male and female cohorts. Pairwise correlation analyses revealed several clinical variables that were significantly correlated with cardiovascular outcomes and were included in covariate-adjusted models. Polygenic risk scores for stroke were significant contributors to model performance in covariate-adjusted models for females, with odds ratios of approximately 1.8. Additional models considering known risk factors, including high cholesterol, diabetes, and age, also demonstrated moderate to strong performance, particularly for myocardial infarction and high blood pressure. Sex-specific differences in model performance were observed for vascular disease, with male models exhibiting stronger performance and a significantly higher odds ratio than the female cohort. Variants used in risk score construction were sex- and outcome-specific, with associations being made during variant filtering procedures.[DL3] Our workflow provides an in-depth analysis of machine-learning-based variant selection for polygenic risk score construction, as well as clinical covariate selection to improve modelling. Modelling procedures that combine genomic and phenotypic variables can provide accurate predictions of cardiovascular disease.