Evaluation of chromosome-level structural variation for acute myocardial infarction risk prediction

Abstract Acute myocardial infarction (AMI) has substantial heritability. Genome-wide association studies (GWAS) often use single-nucleotide polygenic risk scores to estimate this inherited risk, but macro-scale structural alterations remain under-explored. We evaluated whether chromosome-level structural variation (CLSV) provides a predictive signal for AMI. We quantified this variation as statistical moments of microarray Log R Ratio (LRR) distributions across 128 segments of the genome. We performed 1:1 matching between cases and controls on sex at birth, self-reported race, and age (\(\:\le\:\pm\:2\) years) within the All of Us Research Program Controlled Tier dataset (\(\:N=3,290\)) to eliminate demographic confounding. We evaluated Random Forest models across 10 independent stratified cross-validation splits. The CLSV model achieved a 10-split mean held-out test ROC AUC of 0.529 (SD: \(\:0.048\), 10-split range: \(\:0.439\) to \(\:0.604\)). This performance is statistically indistinguishable from a scrambled target control (\(\:0.491\), 10-split range: \(\:0.465\) to \(\:0.546\)) and overlaps random chance (\(\:0.500\)). A positive biological control predicting sex at birth using the same genomic moment extraction pipeline achieved near-perfect discrimination (mean AUC = \(\:0.999\), 10-split range: \(\:0.997\) to \(\:1.000\)), confirming high technical validity. We establish with 95% statistical confidence that any true predictive signal carried by broad structural genomic variation in an unconfounded matched cohort is less than an upper 95% confidence limit of 0.585 (Nadeau–Bengio corrected, mean AUC = 0.529). Our findings suggest that macro-scale structural variation does not explain missing heritability or provide clinical risk discrimination for AMI.