Identifying Interaction Effects of Single Nucleotide Polymorphisms (SNPs) on Disease Risk

This study examines the interaction effects of Single Nucleotide Polymorphisms (SNPs) on hypertension risk, using data from the Framingham Heart Study (FHS): three generations in Massachusetts, United States. Genomic data still represent a significant portion of the body of literature regarding genetic associations with disease risk, yet applications in which machine learning techniques are used for detecting SNP interactions have been researched relatively little. This sets the stage for our research, which is an attempt to address this issue from multiple angles: 1) we have adopted a broad methodology involving stringent quality control, using principal components (PCs) computed on pairwise genetic distances—a PCA-based correction method that can reduce Type I errors while not supplanting allelic association analysis to account for population structure, and an imputation approach based on K-nearest neighbors to render more plausible biological interpretations of results. Our analysis is further refined with the removal of SNPs that either fall below Minor Allele Frequency (MAF) thresholds or fail to meet Hardy-Weinberg Equilibrium requirements, as well as a multicollinearity assessment of predictors aimed at confirming the reliability of our findings. Results provide evidence of significant SNP-SNP interactions influencing hypertension risk, offering new insights into the genetic architecture of the disease. In addition to providing fundamental knowledge on SNP interactions in disease susceptibility, this study exemplifies the utility of machine learning approaches for genomic research and opens an avenue toward more comprehensive studies on genetic risk factors underlying health and disease. We incorporated a simulation model to assess SNP interaction effects on disease risk in a controlled setting. The simulation begins by generating genotype data for n individuals and m SNPs, where the Minor Allele Frequencies (MAFs) are modeled using a Beta distribution. Genotypes are simulated under Hardy-Weinberg Equilibrium (HWE) conditions using a multinomial distribution, and disease status is generated using a logistic model with main and interaction effects. This simulated approach complements the real data study, offering a robust framework to validate SNP interactions and assess the impact of genetic variations on disease susceptibility.