Statistical Analysis for Genome Data Based on Multiple SNPs Using Kernel Machine Based Test

This study explores a nonparametric regression method to detect significant associations in genome-wide data using multi-marker tests. Data from the Wellcome Trust Case Control Consortium (WTCCC) on Crohn's disease were analyzed., utilizing a Gaussian kernel with various bandwidths and thresholds. The effectiveness of the proposed method was compared against the Bonferroni correction method. Simulations were conducted using genotype data from the 1958 British Birth Cohort., focusing on disease SNPs in low and moderate linkage disequilibrium (LD). The true positive (TP) and false positive (FP) rates were assessed for different gene effect sizes. Results indicated that the Gaussian kernel with a threshold of 3 provided the best overall performance., yielding higher TP rates and comparable FP rates to the Bonferroni correction. Additionally., the method was evaluated under the assumption that disease is caused by the joint action of SNPs within a pathway., showing superior performance for low LD and moderate LD conditions. The proposed nonparametric regression approach effectively identifies significant SNP associations., demonstrating its potential for genome-wide association studies (GWAS) by offering improved TP rates while maintaining controlled FP rates. These findings suggest that the Gaussian kernel method could be a valuable tool for genetic epidemiology., providing a robust alternative to traditional correction methods.