Pedigree-based random effect tests to screen gene pathways

Abstract

The new generation of sequencing platforms opens new horizons in the genetics field. It is possible to exhaustively assay all genetic variants in an individual and search for phenotypic associations. The whole genome sequencing approach, when applied to a large human sample like the San Antonio Family Study, detects a very large number (>25 million) of single nucleotide variants along with other more complex variants. The analytical challenges imposed by this number of variants are formidable, suggesting that methods are needed to reduce the overall number of statistical tests. In this study, we develop a single degree-of-freedom test of variants in a gene pathway employing a random effect model that uses an empirical pathway-specific genetic relationship matrix as the focal covariance kernel. The empirical pathway-specific genetic relationship uses all variants (or a chosen subset) from gene members of a given biological pathway. Using SOLAR's pedigree-based variance components modeling, which also allows for arbitrary fixed effects, such as principal components, to deal with latent population structure, we employ a likelihood ratio test of the pathway-specific genetic relationship matrix model. We examine all gene pathways in KEGG database gene pathways using our method in the first replicate of the Genetic Analysis Workshop 18 simulation of systolic blood pressure. Our random effect approach was able to detect true association signals in causal gene pathways. Those pathways could be easily be further dissected by the independent analysis of all markers.

Finding potential cis-regulatory loci using allele-specific chromatin accessibility as weights in a kernel-based variance component test

“…The first two principal components (PC1, PC2) (estimated as described in Peralta et al [7] and Almeida et al [8]), were added to account for any unknown population substructure that might be present.…”

“…We used the variance component model previously described in Peralta et al [7] and Almeida et al [8], in conjunction with the nonweighted and FP-weighted covariance kernels derived from the SNP dosages described above, to estimate the proportion of the phenotypic variance, h geff 2 , explained by allele-specific genetic variants found within DHSs in an unrelated CEU-CEPH individual. The h geff 2 variance component, and its significance, was estimated for each real and simulated expression phenotype using SOLAR, a flexible genetic variance component analysis program with a focus on general pedigrees [14].…”

“…We have also developed a variance component based burden test to determine the contribution of localized relationship kernels to the trait variance [7, 8]. Here, we test if by combining both lines of research we could detect potential cis -acting regulatory loci.…”

A Kernel of Truth

“…This idea could be extended to computing heritabilities on a chromosomal or a gene segment basis (Yang et al, 2011a). One could also leverage this approach to compute the K relevant for a metabolic pathway and to estimate pathway-specific variance components (Almeida et al, 2012). We have shown that it is possible to accurately recover both total and local (i.e., QTL-specific) heritability estimates by using only empirical GRKs in a known pedigree situation (Day-Williams et al, 2011).…”

Independent test assessment using the extreme value distribution theory

“…Large genetic and epidemiologic projects are beginning to utilize whole genome sequencing (WGS) to identify genetic variants (especially novel rare variants) that could explain the missing heritability paradox [1]. The paradox had been formulated during the genome wide association (GWA) era; those platforms are based on the CDCV (common disease common variant) premise.…”