Integrating multiple genomic data: sparse representation based biomarker selection for blood pressure

Abstract

BackgroundAlthough many genes have been implicated as hypertension candidates, to date, few studies have integrated different types of genomic data for the purpose of biomarker selection.MethodsApplying a newly proposed sparse representation based variable selection (SRVS) method to the Genetic Analysis Workshop19 data, we analyzed a combined data set consisting of 11522 gene expressions and 354893 single-nucleotide polymorphisms (SNPs) from 397 subjects (case/control: 151/246), with the aim to identify potential biomarkers for blood pressure using both gene expression measures and SNP data.ResultsAmong the top 1000 variables (SNPs/gene expressions = 575/425) selected, the bioinformatics analysis showed that 302 were plausibly associated with blood pressure. In addition, we identified 173 variables that were associated with body weight and 84 associated with left ventricular contractility. Together, 55.9 % of the top 1000 variables showed associations with blood pressure related phenotypes(SNP/gene expression =348/211).ConclusionsOur results support the feasibility of the SRVS algorithm in integrating multiple data sets of different structure for comprehensive analysis.

Joint analysis of multiple phenotypes: summary of results and discussions from the Genetic Analysis Workshop 19

“…This can then be followed by more detailed questions as to how the multiple phenotypes are analyzed statistically, and whether, for example, they are considered as dependent outcome variables or as covariates on the same level as genotypes. Common to all contributions of this working group [ 8 – 11 ] was the search for functional single nucleotide variants (SNVs) influencing the blood pressure phenotypes, with different motivations and approaches for integrating multiple phenotypes into the analysis. SBP and DBP show a high correlation, with correlation coefficients between 0.5 and 0.8, depending on the adjustment applied for covariates.…”

“…However, 1 group applied a log-transformation [ 11 ] to BP; 2 groups [ 8 , 11 ] applied standard adjustments for the nongenetic covariates age, sex, and smoking; and the other 2 groups [ 9 , 10 ] adjusted BP for the effect of antihypertensive medication and other nongenetic covariates using a censored regression model [ 15 ]. Furthermore, 1 group [ 10 ] looked at longitudinal effects, and 2 groups [ 8 , 9 ] considered SBP only but included gene expression measures, which made a comparison of identified SNVs difficult. Table 1 provides a brief overview of the analyzed samples and data, and shows the employed statistical models, implementation, and main findings of all contributions.…”

“…These models were then fitted sequentially for different SNVs or sets of SNVs with copula models [ 9 ], structural equation models (SEM) [ 10 ], or multivariate linear mixed effect regression [ 11 ]. As another novel approach to the high dimensionality, Cao et al [ 8 ] used a sparse representation based variable selection (SRVS) to extract relevant genes based on signatures from the entire data, including all SNVs and all gene expressions. In this regard, all approaches could be described as different forms of meta-dimensional approaches, with either using a sparse statistical model and analyzing all the SNP and gene expression data simultaneously, or by posing hypothesis-based restrictions and testing triplets of SBP, DBP, and SNVs or of SBP, gene expression, and SNVs sequentially.…”

“…In this regard, all approaches could be described as different forms of meta-dimensional approaches, with either using a sparse statistical model and analyzing all the SNP and gene expression data simultaneously, or by posing hypothesis-based restrictions and testing triplets of SBP, DBP, and SNVs or of SBP, gene expression, and SNVs sequentially. In more detail, Cao et al [ 8 ] analyzed 11,522 gene expression measures, 354,893 SNPs and 6 nongenetic variables simultaneously. Konigorski et al [ 9 ] and Song et al [ 10 ], each tested models incorporating multiple phenotypes (SBP and 1 gene expression measure [ 9 ], or 3 longitudinal SBP and 3 DBP measures [ 10 ]) together with 1 SNV/SNP, sequentially for 68,727 SNVs [ 9 ] and 460,359 SNPs [ 10 ].…”

“…As another view on the approaches, Konigorski et al [ 9 ], Song et al [ 10 ], and Sun et al [ 11 ] applied new methodologies and implementations for models of multiple dependent outcome measures to integrate BP and gene expression as phenotypes in the analysis. On the other hand, Cao et al [ 8 ] looked at gene expression at the same level as genotypes to derive joint signatures relevant for BP, and applied a new approach to extend the literature on how to integrate and mine multiple covariates for interesting structures. Hence, the approaches focus on slightly different aspects of transcriptional and translational processes.…”