Search

Gene set-based signal detection analyses are used to detect an association between a trait and a set of genes by accumulating signals across the genes in the gene set. Since signal detection is concerned with identifying whether any of the genes in the gene set are non-null, a goodness-of-fit (GOF) test can be used to compare whether the observed distribution of gene-level tests within the gene set agrees with the theoretical null distribution. Here, we present a flexible gene set-based signal detection framework based on tail-focused GOF statistics. We show that the power of the various statistics in this framework depends critically on two parameters: the proportion of genes within the gene set that are non-null and the degree of separation between the null and alternative distributions of the gene-level tests. We give guidance on which statistic to choose for a given situation and implement the methods in a fast and user-friendly R package, wHC (https://github.com/mqzhanglab/wHC). Finally, we apply these methods to a whole exome sequencing study of amyotrophic lateral sclerosis.

Purpose: Design and characterization of a radiation biodosimetry device are complicated by the fact that the requisite data are not available in the intended use population, namely humans e...

Background Time- and dose-to-event phenotypes used in basic science and translational studies are commonly measured imprecisely or incompletely due to limitations of the experimental design or data collection schema. For example, drug-induced toxicities are not reported by the actual time or dose triggering the event, but rather are inferred from the cycle or dose to which the event is attributed. This exemplifies a prevalent type of imprecise measurement called grouped failure time, where times or doses are restricted to discrete increments. Failure to appropriately account for the grouped nature of the data, when present, may lead to biased analyses. Results We present groupedSurv, an R package which implements a statistically rigorous and computationally efficient approach for conducting genome-wide analyses based on grouped failure time phenotypes. Our approach accommodates adjustments for baseline covariates, and analysis at the variant or gene level. We illustrate the statistical properties of the approach and computational performance of the package by simulation. We present the results of a reanalysis of a published genome-wide study to identify common germline variants associated with the risk of taxane-induced peripheral neuropathy in breast cancer patients. Conclusions groupedSurv enables fast and rigorous genome-wide analysis on the basis of grouped failure time phenotypes at the variant, gene or pathway level. The package is freely available under a public license through the Comprehensive R Archive Network. Electronic supplementary material The online version of this article (10.1186/s12859-019-2899-x) contains supplementary material, which is available to authorized users.

Background In order to better understand complex diseases, it is important to understand how genetic variation in the regulatory regions affects gene expression. Genetic variants found in these regulatory regions have been shown to activate transcription in a tissue-specific manner. Therefore, it is important to map the aforementioned expression quantitative trait loci (eQTL) using a statistically disciplined approach that jointly models all the tissues and makes use of all the information available to maximize the power of eQTL mapping. In this context, we are proposing a score test-based approach where we model tissue-specificity as a random effect and investigate an overall shift in the gene expression combined with tissue-specific effects due to genetic variants. Results Our approach has 1) a distinct computational edge, and 2) comparable performance in terms of statistical power over other currently existing joint modeling approaches such as MetaTissue eQTL and eQTL-BMA. Using simulations, we show that our method increases the power to detect eQTLs when compared to a tissue-by-tissue approach and can exceed the performance, in terms of computational speed, of MetaTissue eQTL and eQTL-BMA. We apply our method to two publicly available expression datasets from normal human brains, one comprised of four brain regions from 150 neuropathologically normal samples and another comprised of ten brain regions from 134 neuropathologically normal samples, and show that by using our method and jointly analyzing multiple brain regions, we identify eQTLs within more genes when compared to three often used existing methods. Conclusions Since we employ a score test-based approach, there is no need for parameter estimation under the alternative hypothesis. As a result, model parameters only have to be estimated once per genome, significantly decreasing computation time. Our method also accommodates the analysis of next- generation sequencing data. As an example, by modeling gene transcripts in an analogous fashion to tissues in our current formulation one would be able to test for both a variant overall effect across all isoforms of a gene as well as transcript-specific effects. We implement our approach within the R package JAGUAR, which is now available at the Comprehensive R Archive Network repository. Electronic supplementary material The online version of this article (doi:10.1186/s12859-016-1123-5) contains supplementary material, which is available to authorized users.

The case-control study is a common design for assessing the association between genetic exposures and a disease phenotype. Though association with a given (case-control) phenotype is always of primary interest, there is often considerable interest in assessing relationships between genetic exposures and other (secondary) phenotypes. However, the case-control sample represents a biased sample from the general population. As a result, if this sampling framework is not correctly taken into account, analyses estimating the effect of exposures on secondary phenotypes can be biased leading to incorrect inference. In this paper, we address this problem and propose a general approach for estimating and testing the population effect of a genetic variant on a secondary phenotype. Our approach is based on inverse probability weighted estimating equations, where the weights depend on genotype and the secondary phenotype. We show that, though slightly less efficient than a full likelihood-based analysis when the likelihood is correctly specified, it is substantially more robust to model misspecification, and can out-perform likelihood-based analysis, both in terms of validity and power, when the model is misspecified. We illustrate our approach with an application to a case-control study extracted from the Framingham Heart Study. Copyright © 2016 John WileySons, Ltd.

Compound heterozygous mutations are mutations that occur on different copies of genes and may completely “knock-out” gene function. Compound heterozygous mutations have been implicated in a large number of diseases, but there are few statistical methods for analyzing their role in disease, especially when such mutations are rare. A major barrier is that phase information is required to determine whether both gene copies are affected and phasing rare variants is difficult. Here, we propose a method to test compound heterozygous and recessive disease models in case–parent trios. We propose a simple algorithm for phasing and show via simulations that tests based on phased trios have almost the same power as tests using true phase information. A further complication in the study of compound heterozygous mutations is that only families where both parents carry mutations are informative. Thus, the informative sample size will be quite small even when the overall sample size is not, making asymptotic approximations of the null distribution of the test statistic inappropriate. To address this, we develop an exact test that will give appropriate P-values regardless of sample size. Using simulation, we show that our method is robust to population stratification and significantly outperforms other methods when the causal model is recessive.