DEBrowser: interactive differential expression analysis and visualization tool for count data
Sequencing data has become a standard measure of diverse cellular activities. For example, gene expression is accurately measured by RNA sequencing (RNA-Seq) libraries, protein-DNA interactions are captured by chromatin immunoprecipitation sequencing (ChIP-Seq), protein-RNA interactions by crosslinking immunoprecipitation sequencing (CLIP-Seq) or RNA immunoprecipitation (RIP-Seq) sequencing, DNA accessibility by assay for transposase-accessible chromatin (ATAC-Seq), DNase or MNase sequencing libraries. The processing of these sequencing techniques involves library-specific approaches. However, in all cases, once the sequencing libraries are processed, the result is a count table specifying the estimated number of reads originating from each genomic locus. Differential analysis to determine which loci have different cellular activity under different conditions starts with the count table and iterates through a cycle of data assessment, preparation and analysis. Such complex analysis often relies on multiple programs and is therefore a challenge for those without programming skills.
We developed DEBrowser as an R bioconductor project to interactively visualize every step of the differential analysis, without programming. The application provides a rich and interactive web based graphical user interface built on R’s shiny infrastructure. DEBrowser allows users to visualize data with various types of graphs that can be explored further by selecting and re-plotting any desired subset of data. Using the visualization approaches provided, users can determine and correct technical variations such as batch effects and sequencing depth that affect differential analysis. We show DEBrowser’s ease of use by reproducing the analysis of two previously published data sets.
DEBrowser is a flexible, intuitive, web-based analysis platform that enables an iterative and interactive analysis of count data without any requirement of programming knowledge.
KeywordsDifferential expression Data visualization Interactive data analysis
Assay for transposase-accessible chromatin
Chromatin immunoprecipitation sequencing
Cross-linking immunoprecipitation sequencing
Human monocyte derived mouse dendritic cells
- JNK1 and JNK1
Jun terminal kinases
JNK1 and JNK2 double knockout
Median ratio normalization
Principal component analysis
RNA immunoprecipitation sequencing
Relative Log Expression
Trimmed Mean of M-values
Data assessment evaluates the impact of latent factors that may not be related to biological differences. Such differences might come from technical factors, such as DNA/RNA fragmentation, the number of PCR cycles, or sequencing depth that may altogether confound the actual biologically relevant differences present in different samples [16, 17]. Therefore, the count matrix represents a combination of both biological and technical variation. Unsupervised methods such as clustering and principal component analysis (PCA) are powerful ways to identify technical sources of variance [18, 19, 20]. Data preparation builds on data assessment to determine and then apply the best approaches to reduce the impact of unwanted sources of variance. Typically this includes the elimination of low quality samples, and filtering or removal of features having low counts [21, 22, 23, 24, 25]. Data preparation may also include batch effect correction , which removes variability between groups of samples resulting from technical differences in the day, reagents or experimental personnel involved with sample preparation. Following these pre-processing steps often leads to more accurate differential analysis. Differential Analysis seeks to identify features that have higher or lower counts between groups of samples that represent different biological conditions.
We created DEBrowser to enable iterative analysis by non-programmers to achieve similar results to those obtained by investigators well versed in the R programming language. DEBrowser facilitates a modular iterative analysis and visualization process through an intuitive user interface, integrating multiple algorithms and visualization techniques. The goal is to allow users to iteratively inspect and apply each of the many approaches comprised of the three stages described above. DEBrowser provides an evaluation of the results at each analysis step and determines whether further improvements are necessary. DEBrowser goes further than providing static plots or heatmaps: it allows users to explore any anomaly or potential result in an interactive and dynamic manner by zooming-in on data subsets and selecting or hovering over any regions or genes of interest to plot a heatmap, bar, or box plots that updates dynamically based on the user’s selection. For example, users may select the most significantly differentially expressed genes on a volcano plot, and re-display them in a heatmap, that can be further used to inspect the expression of each individual genes across all replicates.
DE analysis and visualization packages
A number of graphical user interfaces address the need for user-friendly, programming-free visualizations [27, 28, 29, 30, 31, 32, 33] (Additional file 1: Table S1). However, all of these approaches have limited interactivity for users to carry out more sophisticated analysis. Similar to DEBrowser, existing tools accept count data as input to visualize, identify, and perform differential analysis and gene ontology. They also visualize results using scatter, MA or volcano plots, as well as heatmaps and PCA plots. DEBrowser goes one step further by enabling hands-on manipulation of the data and by enabling users to re-plot selected subsets of data. These features make DEBrowser a sophisticated tool for data exploration. DEBrowser allows users to color genes that exceed the different fold change cutoffs after differential expression analysis with only a few clicks. Furthermore, all plots are immediately redrawn immediately after changes in the plotting parameters or after any data subsetting operation. Similarly, for easy access to the underlying data, DEBrowser supports hovering to obtain detailed information of individual data points.
In contrast to existing visualization and differential analysis tools, DEBrowser places a strong focus on data assessment and preparation: it intrinsically supports normalization and batch correction methods. Once the user identifies a clear bias in data preparation, processing or sequencing, DEBrowser allows the user to minimize technical variation between samples using proven methods [26, 34, 35, 36, 37, 38]. This capability is intended to address the needs of large projects that process samples over an extended period of time, and to help users compare samples available from public repositories originating from different laboratories.
DEBrowser is implemented in R as a shiny application using generic shiny components and layouts . In particular, DEBrowser relies on R’s plotly package  both for interactive plots and to display multi-panel data.
Shiny’s reactive programming model is used to update plots upon detection of changes in any input control, plot, or any other bound object. Automatic re-plotting reduces the number of clicks required and therefore improves user experience. When a change in a plot parameter is detected, plots bound to that input are redrawn. As a result, with few clicks, users can, for example, change the highlighted genes from the DE results that exceed a significance of 0.05 and a 2-fold change, to those exceeding a significance of 0.01 and a 10-fold change.
In a typical DEBrowser session users upload a count table obtained by processing of sequencing reads and a table specifying sample information: Conditions and batches. No other inputs are needed (Additional file 2: Figure S1).
Design and key features
To show the general applicability of DEBrowser on “count data” from different data types; we used a large data set that we recently generated to study gene regulation in innate immune cells (human monocyte derived dendritic cells, hMDDCs) in response to Toll-like receptor signaling . This study generated RNA-Seq, ChIP-Seq and ATAC-Seq  to track changes in transcription and regulatory element activity in the course of Toll-like receptor (TLR) signaling. We reprocessed all raw sequence reads either as described in the original publication or using more recent approaches (Additional file 3).
These data are ideal to showcase the main features of DEBrowser and how DEBrowser can be used throughout the analysis cycle. Indeed, we show how DEBrowser was used for data assessment to identify batch effects, and data preparation by filtering low count features and removing batch effects and then performing differential analysis.
Quality control (QC): Quality control of the count data is a fundamental step in analysis, yet it is not well supported in current applications. With DEBrowser users can easily establish whether normalization, batch correction or sample removal are necessary, or if the data is suitable for differential analysis. To this end, DEBrowser implements PCA, all2all scatter, heatmaps, interquartile range (IQR) and density plots of each sample. These plots can be drawn using a user defined subset of genes, for example by choosing the top N variable genes as defined by coefficient of variance [40, 41]. The subset of genes can be defined graphically by either an expression cutoff or by directly selecting them from another plot.
Plots available for data assessment
Principal component analysis (PCA)
PCA finds an ordered coordinate transformation whose basis capture, in decreasing order, the most variance in the data. DEBrowser allows users to plot any pair of principal components in a scatter plot. Once users specify sample information (e.g. condition), DEBrowser uses colors or shapes to group samples. These plots are ideal to detect outliers or batch effects (Fig. 2).
All2all scatter plots
Gives an overview of sample similarities and variance by plotting all pairwise scatter plots (Additional file 2: Figure S2). Low correlation or high variance across replicates will negatively impact the power to detect DE.
IQR and density plots
Interquartile range and density plots display a sample’s quantification distribution in different ways. Using these plots, users can detect any global discrepancy across samples and evaluate the impact of normalization on the distribution of counts. DEBrowser simplifies comparisons by providing plots for both normalized and unnormalized data. Plots are re-drawn as soon as users change the normalization method.
Removal of low coverage features
The count data originating from a sequencing experiment is affected by sequencing depth as well as from differences in the composition of the detected features [44, 45]. DEBrowser supports normalization methods specifically designed for count data: median ratio normalization (MRN) implemented in DESeq2 package [43, 46], Trimmed Mean of M-values (TMM), Relative Log Expression (RLE), and upper quartile methods implemented in the EdgeR package . To evaluate the effect of normalization, DEBrowser immediately displays PCA, IQR and density plots after normalization.
Batch effect correction
When quality control shows a clear batch effect that can be traced back to a technical artifact (e.g. different sequencing devices, different personnel, library kits, reagent batch), DEBrowser allows the users, if the experimental design allows batch correction, to minimize the batch effect . The users can specify a batch for every sample via a simple tab separated file that can be created using a text editor or spreadsheet software. Given a batch specification, DEBrowser supports two different methods: ComBat [26, 49] and Harman . The resulting, batch corrected dataset can be evaluated using the same tools available for initial quality control (Figs. 2b, 3b). DEBrowser provides a platform to detect, correct and evaluate the result of batch correction.
Differential expression (DE) analysis
To demonstrate a typical usage of DEBrowser, we applied DEBrowser on a data set from a previously published study on the role of Jun terminal kinases (JNK1 and JNK2) in the liver and their role in insulin resistance . For this purpose, the authors relied on four different mouse genotypes: wild type (WT), and hepatocyte specific knockouts of Jnk1 and Jnk2 independently (LΔ1, LΔ2), and a double knockout (LΔ1,2). Each genotype was fed either a regular or a high fat diet (HFD). Thereafter, hepatic expression was assayed for each genotype fed with corresponding diet in triplicate using RNA-Seq, resulting in a total of 24 libraries. This study is ideal for DE analysis as it included three replicates per condition. Therefore, we used RSEM  for library quantification and DEBrowser to analyze the resulting read count table.
We then compared the effect of both normal and HFD on all four genotypes. To do so we performed pairwise comparisons between all conditions and selected all genes with a |log2foldChange| > 1 and padj< 0.01 in at least one comparison to display in a heatmap similar to Additional file 2: Figure S3-A in the original report (Fig. 6c). Finally, we used DEBrowser to reproduce Additional file 2: Figure S3-B in original report. To do so we performed gene ontology analysis of the DE genes between WT and LΔ1,2 on HFD, selected genes that are annotated as part of the PPARα pathway and visualized them as a heatmap (Fig. 6d).
It is important to note that the original publication used much older DE methods . When we applied DESeq2 to replicate the analysis we found that it had much greater power to detect differentially expressed genes and indeed at a similar threshold many more genes are called DE. Nevertheless, there is a very good agreement (73%) in the calls made by both methods (Additional file 2: Figure S4), and most importantly, there are no differences with the Gene Ontology enrichments reported in the original publication (Additional file 2: Figure S5).
Gene ontology, disease and pathway discovery
Comparison to related applications
There are several applications, with varying functionalities, available for the exploration and analysis of DE. Most notable ones are, OASIS , VisRseq , DEGUST , DEIVA , WebMeV , Chipster , and DEapp . A comparison of DEBrowser features to those applications is shown in Additional file 1: Table S1.
DEBrowser modular design
To reduce the code complexity and manage the program easier the components were designed in a modular fashion, so that while DEBrowser grows larger, it is easy to build on top of the simple modules. To this end, bar, box and scatter plots, heatmaps modules could be reused multiple times in DEBrowser. We also shared example shiny applications that use individual modules. This modularity increased our development, test speed, and code reusability. For example, the size and the margins of the plots are controlled within the same module in all the plots in DEBrowser. This modular design allows other users to repurpose any of the tools built into DEBrowser for their own packages.
Existing tools do not fully support the full process of differential expression analysis and visualization. Additionally, the plots are usually static and do not allow interactivity to understand the different parts of the data using different parameters reducing the efficiency of data exploration.
In contrast, the DEBrowser application provides users, who do not have any programming experience, the ability to perform their own analysis in an iterative and interactive process that responds dynamically to user inputs. DEBrowser leverages open source components that are in active development in bioconductor [55, 56], thus it benefits from a large community of developers. Its modular design makes it easy to swap components shall new paradigms or projects emerge that provide more ideal functionality than currently available. Therefore, it fills a much-needed void in graphical user interfaces for the analysis of count data that is typical of sequencing assays.
We would like to thank Kyle Gellatly, Pranitha Vangala, Elisa Donnard, Rachel Murphy, Alan Derr, Laney Zuerlein, Stephen McGregor, Michael P. Czech, Leonardo Collado-Torres, and all members of the Garber Lab for their suggestions and comments.
This work was supported by fund from the National Human Genome Research Institute NHGRI grant #U01 HG007910–01 and the National Center for Advancing Translational Sciences grant #UL1 TR001453–01 (M.G).
Availability of data and materials
Project name: DEBrowser
Operation systems: Platform independent
Programming language: R (> = 3.5.0)
Restrictions to use by non-academics: None
AK and MJM AK implemented the package and wrote the manuscript. OY and DO interpreted data. MG supervised the project and wrote the manuscript. AK, MJM and MG conceived the project. All authors have read and approved the final manuscript.
Ethics approval and consent to participate
Consent for publication
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
- 14.Liu T. Use model-based analysis of ChIP-Seq (MACS) to analyze short reads generated by sequencing protein–DNA interactions in embryonic stem cells. In: Methods in Molecular Biology; 2014. p. 81–95.Google Scholar
- 29.Powell DR. Degust: Visualize, explore and appreciate RNA-seq differential gene-expression data; 2015.Google Scholar
- 32.Kallio MA, Aleksi Kallio M, Tuimala JT, Hupponen T, Klemelä P, Gentile M, et al. Chipster: user-friendly analysis software for microarray and other high-throughput data. BMC Genomics 2011;12. doi: https://doi.org/10.1186/1471-2164-12-507
- 33.Li Y, Andrade J. DEApp: an interactive web interface for differential expression analysis of next generation sequence data. Source Code Biol Med. 2017;12. https://doi.org/10.1186/s13029-017-0063-4.
- 40.Sievert C, Parmer C, Hocking T, Chamberlain S, Ram K, Corvellec M, et al. plotly: Create Interactive Web Graphics via “plotly. js.” R package version, vol. 3; 2016.Google Scholar
- 41.Donnard E, Vangala P, Afik S, McCauley S, Nowosielska A, Kucukural A, et al. Comparative analysis of immune cells reveals a conserved regulatory lexicon. Cell Syst. 2018. https://doi.org/10.1016/j.cels.2018.01.002.
- 49.Johnson WE, Evan Johnson W, Li C. Adjusting batch effects in microarray experiments with small sample size using empirical bayes methods. In: Batch effects and noise in microarray experiments, Biostatistics. 2007;8(1):118–27.Google Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.