Unifying the analysis of high-throughput sequencing datasets: characterizing RNA-seq, 16S rRNA gene sequencing and selective growth experiments by compositional data analysis

Abstract

BackgroundExperimental designs that take advantage of high-throughput sequencing to generate datasets include RNA sequencing (RNA-seq), chromatin immunoprecipitation sequencing (ChIP-seq), sequencing of 16S rRNA gene fragments, metagenomic analysis and selective growth experiments. In each case the underlying data are similar and are composed of counts of sequencing reads mapped to a large number of features in each sample. Despite this underlying similarity, the data analysis methods used for these experimental designs are all different, and do not translate across experiments. Alternative methods have been developed in the physical and geological sciences that treat similar data as compositions. Compositional data analysis methods transform the data to relative abundances with the result that the analyses are more robust and reproducible.ResultsData from an in vitro selective growth experiment, an RNA-seq experiment and the Human Microbiome Project 16S rRNA gene abundance dataset were examined by ALDEx2, a compositional data analysis tool that uses Bayesian methods to infer technical and statistical error. The ALDEx2 approach is shown to be suitable for all three types of data: it correctly identifies both the direction and differential abundance of features in the differential growth experiment, it identifies a substantially similar set of differentially expressed genes in the RNA-seq dataset as the leading tools and it identifies as differential the taxa that distinguish the tongue dorsum and buccal mucosa in the Human Microbiome Project dataset. The design of ALDEx2 reduces the number of false positive identifications that result from datasets composed of many features in few samples.ConclusionStatistical analysis of high-throughput sequencing datasets composed of per feature counts showed that the ALDEx2 R package is a simple and robust tool, which can be applied to RNA-seq, 16S rRNA gene sequencing and differential growth datasets, and by extension to other techniques that use a similar approach.

Excess False Positive Rates in Methods for Differential Gene Expression Analysis using RNA-Seq Data

“…Of course, many other methods exist, including but not limited to: Cuffdiff (Trapnell et al (2010)), Cuffdiff2 (Trapnell et al (2013)), NBPSeq (Di, Schafer, Cumbie, and Chang (2011)), TSPM (Auer and Doerge (2011)), baySeq (Hardcastle and Kelly (2010)), EBSeq (Leng et al (2013)), NOISeq (Tarazona, García-Alcalde, Dopazo, Ferrer, and Conesa (2011)), SAMseq (J. Li and Tibshirani (2013)), ShrinkSeq (Van De Wiel et al (2012)), DEGSeq (Wang, Feng, Wang, Wang, and Zhang (2010)), BBSeq (Y.-H. Zhou, Xia, and Wright (2011)), FDM (Singh et al (2011)), RSEM (B. Li and Dewey (2011)), Myrna (Langmead, Hansen, and Leek (2010)), PANDORA (Moulos and Hatzis (2014)), ALDEx2 (Fernandes et al (2014)), PoissonSeq (J. Li, Witten, Johnstone, and Tibshirani (2011)), and GPSeq (Srivastava and Chen (2010)). We provide code that can be easily adapted to any method that runs in R and applied to the publicly available data sets we used, as well as others.…”

Linear Association in Compositional Data Analysis

“…Log-ratio analysis of compositional data does not allow null proportions as an argument of a logarithm, thus requiring special treatment of such data (Martín-Fernández, Hron, Templ, Filzmoser, and Palarea-Albaladejo 2015a;Gloor, Macklaim, Pawlowsky-Glahn, and Egozcue 2017). The present contribution is not aimed at discussing procedures and methods for treating these zero count data, although Bayesian estimation methods show some promise (Fernandes et al 2014). Therefore, the number of genera used in the examples has been reduced to 12 by removing all genera with a total count across all samples of less than 5000, or that have zero counts in more than 100 samples.…”

Guidelines to Statistical Analysis of Microbial Composition Data Inferred from Metagenomic Sequencing

“…Here we will describe R packages specifically intended for metagenomic analysis: basic methods commonly used for a comparison of two or more groups [on the example of their implementation in ALDEx2 package (Fernandes et al, 2014)] and advanced approaches based on generalized linear models allowing both continuous and discrete factors [metagenomeSeq (Paulson et al, 2013), edgeR (McCarthy et al, 2012), DESeq2 (Love et al, 2014), MaAsLin, shotgunFunctionalizeR (Kristiansson et al, 2009)]. Finally, the methods for vector-wise rather than component-wise comparison will be introduced [HMP (La Rosa et al, 2012), vegan (Oksanen et al, 2012), micropower (Kelly et al, 2015)].…”

“…., α n ) is a feature vector (with an additional pseudocount of 0.5 added to each component) and B(α) is the multivariate beta function. The greater the taxon abundance and the less the whole number of reads for the sample, the greater the variance (Fernandes et al, 2014). Substituting the original feature vector with several random vectors generated from the corresponding Dirichlet distribution leads to a more correct estimation of variance and thus of significance of differences.…”

“…Besides stabilizing the variance, such transformation ensures the proper comparison of two components of the same vector (e.g. which of two species is higher in abundance within a single community) even if low-abundance taxa are excluded from the study (as it is often done) (Fernandes et al, 2014). The transformed abundances may be compared using either Wilcoxon or Welch's tests.…”