Empirical Bayes single nucleotide variant-calling for next-generation sequencing data

One of the fundamental computational problems in cancer genomics is the identification of single nucleotide variants (SNVs) from DNA sequencing data. Many statistical models and software implementations for SNV calling have been developed in the literature, yet, they still disagree widely on real datasets. Based on an empirical Bayesian approach, we introduce a local false discovery rate (LFDR) estimator for germline SNV calling. Our approach learns model parameters without prior information, and simultaneously accounts for information across all sites in the genomic regions of interest. We also propose another LFDR-based algorithm that reliably prioritizes a given list of mutations called by any other variant-calling algorithm. We use a suite of gold-standard cell line data to compare our LFDR approach against a collection of widely used, state of the art programs. We find that our LFDR approach approximately matches or exceeds the performance of all of these programs, despite some very large differences among them. Furthermore, when prioritizing other algorithms' calls by our LFDR score, we find that by manipulating the type I-type II tradeoff we can select subsets of variant calls with minimal loss of sensitivity but dramatic increases in precision.

An empirical Bayes test for allelic-imbalance detection in ChIP-seq

Chromatin immunoprecipitation followed by high-throughput sequencing (ChIP-seq) has enabled discovery of genomic regions enriched with biological signals such as transcription factor binding and histone modifications. Allelic-imbalance (ALI) detection is a complementary analysis of ChIP-seq data for associating biological signals with single nucleotide polymorphisms (SNPs). It has been successfully used in elucidating functional roles of non-coding SNPs. Commonly used statistical approaches for ALI detection are often based on binomial testing and mixture models, both of which rely on strong assumptions on the distribution of the unobserved allelic probability, and have significant practical shortcomings. We propose Non-Parametric Binomial (NPBin) test for ALI detection and for modeling Binomial data in general. NPBin models the density of the unobserved allelic probability non-parametrically, and estimates its empirical null distribution via curve fitting. We demonstrate the advantages of NPBin in terms of interpretability of the estimated density and the accuracy in ALI detection using simulations and analysis of several ChIP-seq data sets. We also illustrate the generality of our modeling framework beyond ALI detection by an application to a baseball batting average prediction problem. This article has supplementary material available at Biostatistics online. The code and the sample input data have been also deposited to github https://github.com/QiZhangStat/ALIdetection.

An empirical Bayes change-point model for identifying 3' and 5' alternative splicing by next-generation RNA sequencing

MOTIVATION

Next-generation RNA sequencing (RNA-seq) has been widely used to investigate alternative isoform regulations. Among them, alternative 3 ': splice site (SS) and 5 ': SS account for more than 30% of all alternative splicing (AS) events in higher eukaryotes. Recent studies have revealed that they play important roles in building complex organisms and have a critical impact on biological functions which could cause disease. Quite a few analytical methods have been developed to facilitate alternative 3 ': SS and 5 ': SS studies using RNA-seq data. However, these methods have various limitations and their performances may be further improved.

RESULTS

We propose an empirical Bayes change-point model to identify alternative 3 ': SS and 5 ': SS. Compared with previous methods, our approach has several unique merits. First of all, our model does not rely on annotation information. Instead, it provides for the first time a systematic framework to integrate various information when available, in particular the useful junction read information, in order to obtain better performance. Second, we utilize an empirical Bayes model to efficiently pool information across genes to improve detection efficiency. Third, we provide a flexible testing framework in which the user can choose to address different levels of questions, namely, whether alternative 3 ': SS or 5 ': SS happens, and/or where it happens. Simulation studies and real data application have demonstrated that our method is powerful and accurate.

AVAILABILITY AND IMPLEMENTATION

The software is implemented in Java and can be freely downloaded from http://ebchangepoint.sourceforge.net/

CONTACT

zhiwei@njit.edu.

Accounting for uncertainty in DNA sequencing data

Science is defined in part by an honest exposition of the uncertainties that arise in measurements and propagate through calculations and inferences, so that the reliabilities of its conclusions are made apparent. The recent rapid development of high-throughput DNA sequencing technologies has dramatically increased the number of measurements made at the biochemical and molecular level. These data come from many different DNA-sequencing technologies, each with their own platform-specific errors and biases, which vary widely. Several statistical studies have tried to measure error rates for basic determinations, but there are no general schemes to project these uncertainties so as to assess the surety of the conclusions drawn about genetic, epigenetic, and more general biological questions. We review here the state of uncertainty quantification in DNA sequencing applications, describe sources of error, and propose methods that can be used for accounting and propagating these errors and their uncertainties through subsequent calculations.

Longitudinal analysis is more powerful than cross-sectional analysis in detecting genetic association with neuroimaging phenotypes

Most existing genome-wide association analyses are cross-sectional, utilizing only phenotypic data at a single time point, e.g. baseline. On the other hand, longitudinal studies, such as Alzheimer's Disease Neuroimaging Initiative (ADNI), collect phenotypic information at multiple time points. In this article, as a case study, we conducted both longitudinal and cross-sectional analyses of the ADNI data with several brain imaging (not clinical diagnosis) phenotypes, demonstrating the power gains of longitudinal analysis over cross-sectional analysis. Specifically, we scanned genome-wide single nucleotide polymorphisms (SNPs) with 56 brain-wide imaging phenotypes processed by FreeSurfer on 638 subjects. At the genome-wide significance level P < 1.8 x 10(9)) or a less stringent level (e.g. P < 10(7)), longitudinal analysis of the phenotypic data from the baseline to month 48 identified more SNP-phenotype associations than cross-sectional analysis of only the baseline data. In particular, at the genome-wide significance level, both SNP rs429358 in gene APOE and SNP rs2075650 in gene TOMM40 were confirmed to be associated with various imaging phenotypes in multiple regions of interests (ROIs) by both analyses, though longitudinal analysis detected more regional phenotypes associated with the two SNPs and indicated another significant SNP rs439401 in gene APOE. In light of the power advantage of longitudinal analysis, we advocate its use in current and future longitudinal neuroimaging studies.