Nonlinear functional regression model for sequencing-based association and gene-gene interaction analysis of physiological traits and their applications to sleep apnea
Quantitative genetics studies usually focus on static quantitative traits. However, more and more studies have interests in functional dynamic quantitative traits . Those traits are changing with time or changing by position. Moreover, for the next sequence genetic data set, we often have millions of variants. Consider that those genetic variants are located in a genomic line. It is natural for us to model those genetic variants as a function of their physical positions. In this study, we proposed models with both functional predictor and functional response. The traditional multivariate analysis fails to apply in this data set due to the high dimensionality. In order to solve this problem, we have developed a Functional Regression Model (FRM). We model the temporal quantitative trait as a function of time, model the genetic variants as a function of the genetic position, and model the genetic additive effect as a bivariate function of both of time and the genomic position.^ From the simulation results, we find that the type I error rates of the FRM is correct. Also, compared with cross-sectional methods, the FRM had the highest power to detect the real association amount the temporal quantitative traits and the genes.^ The proposed method was applied to 833 individuals from Starr County health studies sleep data. The measurement of oxygen saturation is 35,280 seconds over a night. Also, this GWAS data set contains 20,763 genes which included 906,598 SNPs. We found that 65 genes were significantly associated with oxygen saturation functional trait with P-values ranging from 2.4E-6 to 2.5E-21, including gene RELB, which are confirmed in the literature. Among them, 5 genes are related to NF-kappa B signaling pathway, which is related to the pathogenetic mechanisms of obstructive sleep apnea. However, using mean value of oxygen saturation over the whole sleep time period as a quantitative trait, we only identified two genes associated with oxygen saturation. Our results clearly demonstrate that the functional regression model with both functional response and functional predictors substantially outperform the traditional genetic model with scalar trait. For the epistasis analysis, we tested all 148,910,653 possible gene pair from 17,258 genes. We found that 13 pair of genes that interacted associate to the oxygen saturation functional trait. These include 10 genes which are related to the Obesity.^
Lee, Dung-Yang, "Nonlinear functional regression model for sequencing-based association and gene-gene interaction analysis of physiological traits and their applications to sleep apnea" (2015). Texas Medical Center Dissertations (via ProQuest). AAI3721382.