DEvis: an R package for aggregation and visualization of differential expression data

RNA-sequencing (RNA-seq) has provided insights into mechanisms of host response to genetic and infectious diseases, host-pathogen interactions, cancer research, and plant biology [1–6]. Differential expression analysis of RNA-Seq data allows researchers to identify discriminating factors between experimental conditions. To that end, tools for performing differential expression analysis such as DESeq2, EdgeR, and Limma have been widely used in recent years [7–9]. While these tools provide a statistically rigorous framework for the identification of differentially-expressed genes, they require extensive effort on the part of bioinformaticians, who must coerce the resulting data from these packages into different data structures for interpretation or visualization. Furthermore, most experimental designs require complex contrasts that consider multiple factors and time points. This further complicates the analysis and makes identification of the most relevant genes with regard to biological factors or experimental conditions increasingly difficult as the scale of the experiment increases. These problems can be addressed by data aggregation methods that simplify combination and restructuring of transcriptomic data, and provide the flexible visualization tools needed to examine multiple factors of aggregated data, identify patterns of expression, and find transcriptomic profiles that correspond to biological and experimental conditions. Currently, however, few such tools are readily available, and researchers who want to aggregate or visualize their data are responsible for identifying and implementing methods that may be difficult to find, rely on obsolete dependencies, have complex setup procedures, or rely on external software packages that require the export and restructuring of data.

Here, we introduce DEvis, a comprehensive toolset for data aggregation, visual analytics, and exploratory analysis. DEvis addresses the limitations currently inherent to differential expression analysis by making it possible to manipulate and visualize transcriptomic data in easy and flexible ways. This toolset implements a wide range of visualizations for analyzing RNA-seq differential expression results, as well as data aggregation, transformation, sorting, and selection methods that smoothly integrate with the DESeq2 differential expression package, transforming DESeq2 into a more accessible platform for differential expression analysis with convenient visualization and data control facilities. Also, as DEvis, DESeq2, EdgeR and Limma all require similar data structures as input, most researchers can easily plug their input data into DEvis and quickly generate results. DEvis additionally implements a result management system that automatically organizes and stores analysis results, further simplifying the challenge of complex transcriptomic analysis, creation of publication-quality figures and contributing to the reproducibility of analysis.

An example analysis was performed, to validate the results and assess the features of DEvis, from a recently published RNA-Seq study on the pathogenicity of Ebola and Reston viruses in humans [4]. An in-depth version of this analysis is provided in the DEvis vignette document. For the purposes of this paper, however, a simplified explanation is provided as an overview of a straightforward analysis.

Data for time (6 h, 1d, 2d) and treatment (ebov, restv, lps, mock) in human macrophages was downloaded from SRA (SRP078152). Data were analyzed using DEvis and several preprocessing steps were performed before TMM normalization. Differential expression calculations were carried out followed by union-based aggregation of results. The results of merging for each time/virus combination were examined using gene significance density plots that showed strong agreement between significant genes at each time point between restv and ebov, and a subset of genes being significant in either restv or ebov, but not both (Fig. 1i). This was determined to be a valid aggregation for this data set, as these are the same family of viruses and the hypothesis of the original study was that restv was not pathogenic in humans, in which case this level of similarity would be expected. In parallel, two-way hierarchical clustering was performed to examine sample similarity (Fig. 1a), and pre-normalization (not shown) and post-normalization (Fig. 1b) plots were generated to check the integrity of the data, the effects of normalization, and to examine summaries of identified differentially expressed genes (Fig. 1e). Multi-dimensional scaling (MDS) with confidence intervals (Fig. 1c) and MDS with convex hulls (Fig. 1d) were used to analyze data clustering, based on differences in metadata, to identify the factors influencing the sample set as a whole and the subsets of differentially expressed genes. Distributions of expression levels were examined using boxplots (Fig. 1f) and dispersion plots (Fig. 1g), and significance and expression changes were visualized for all contrasts using volcano plots (Fig. 1h). Heat maps were used to investigate the expression of genes across contrasts using several sorting and selection criteria (Fig. 1k) and profiles of the expression changes in subsets of these genes were examined as a function of infection and time (Fig. 1l). Certain genes of interest, as identified in heat map analysis, were examined individually with regard to infection conditions using gene specific boxplots with Wilcoxon tests (Fig. 1j). Co-expression analysis was performed (not shown) and groups of differentially expressed genes with differences in patterns of expression over time were identified for each infection condition. Figures and relevant data in all steps were automatically organized and stored as image and data files.

In this example, each plot can be generated using a single line of code (as shown in the vignette), and exploratory analysis can be performed through the use of different parameters and metadata combinations for each visualization. If a bioinformatician were to perform the same types of analyses as shown here, substantial computations, data selection, and reformatting would be required, consuming time, increasing the potential for human error, and potentially reducing reproducibility.

DEvis provides a user friendly and versatile tool set for aggregation, visualization, and exploratory analysis of transcriptomic data. This package provides a single repository of tools required for the analysis and investigation of complex transcriptomic data sets, as well as automated project management functionality that standardizes analysis, reduces workload, and creates reproducible results. Furthermore, it obliviates the need for maintaining multiple scripts and implementations of various software packages and their dependencies.

Extensive documentation is available for DEvis in the form of a technical manual that outlines the detailed functionality and parameter usage of each method, and a vignette document, which functions as a case-study and tutorial on the usage and caveats of DEvis. This vignette was created based on analysis of data from the previously outlined study on the pathogenicity of Ebola and Reston viruses in humans [4]. The technical documentation and vignettes are available with the DEVis package at https://github.com/price0416/DEvis.

Project name: DEvis.

Project home page: https://github.com/price0416/DEvis

Operating system(s): Platform independent.

License: LGPL.

Programming language: R.

Other requirements: None.

Any restrictions to use by non-academics: LGPL license, open source.