Abstract

Background

Cancer includes a group of largely heterogeneous and complex diseases. The development of massive genomic analysis techniques and the coordinated effort of different international consortiums are providing the research community with comprehensive data of tissue samples from thousands of cancer patients. Free access to these data represents an invaluable opportunity for the research community to get insights into the molecular profiles of cancers allowing, for instance, identification of new candidate markers, in silico hypothesis testing, or validation of in-house results through an independent series.

Cancer research is benefited by coordinated efforts on data acquisition and analysis. The Cancer Genome Atlas (TCGA) [1] is probably the most ambitious of these initiatives offering a multilayered view of genomics and epigenomics data together with clinicopathological information of more than 30 human cancer types. Major contributions of TCGA consortium are illustrated by the opening manuscripts (mainly devoted to report the mutational landscapes) of each one of the considered human cancers [2]; but a large number of studies are taking advantage of the available datasets [3].

TCGA Data Portal [4] is the default gateway to retrieve data offering a text file for each processed sample belonging to a dataset generated using a given platform. Compressed packages for multiple samples are allowed. For instance, it allows downloading 992 files with the clinical data related to samples of the Colon Adenocarcinoma (COAD) dataset, but only provides RNA-Seq results for 303 of them. It is worth noting that as the number of files and their sizes increase, further processing of this information exceeds the spreadsheet-based analysis capabilities and requires bioinformatics skills.

Cancer research teams working on specific aspects of the disease (e.g., exploration of candidate biomarkers, deregulation of signaling pathways, or mechanisms of gene regulation, just to name a few) might be more interested in rapidly querying specific subsets of the data (e.g., expression levels of a named gene or the methylation status of its promoter region) rather than obtaining bulky datasets as offered by the data portal. To facilitate the access and exploitation of TCGA data, several tools have been developed to compare and to explore the different levels of data, facilitating the discovery of associations and correlations, as well as allowing ascertaining the significance of certain alterations in cancer [5–7].

The high level of complexity of transcriptomic and epigenetic profiles in normal and cancer samples poses challenging scenarios to analyze, represent, and interpret both gene-specific and bulk data. Important differences among samples may be easily missed if one-size-fits-all standardized criteria are used. For instance, gene expression differences may appear as alternative transcripts and be missed in summarizing values. Moreover, relevant DNA methylation changes may occur outside predefined promoter domains or just affect a subregion.

Here we present Wanderer, a very simple and intuitive Web tool allowing real-time access and visualization of gene expression and DNA methylation profiles from TCGA data using gene targeted queries. Wanderer is addressed to a broad variety of experimentalists and clinicians without deep bioinformatics skills.

Results

DNA methylation profiling

For any given gene selected by the investigator, the tool provides the detailed individual beta values of all the HumanMethylation450 probes inside or in the vicinity of the gene. Graphs for normal and tumor samples from any of the available TCGA datasets are readily produced (Fig. 1). A summarizing plot is also generated, and statistical comparisons are performed to identify differential DNA methylation between normal and tumor samples at the single probe level. Wanderer performs Wilcoxon rank sum test on normal versus tumor provided there are at least two observations in each group. Although this is readily reported, we note that addressing the significance on few observations and without inspecting the data distribution might be meaningless. Even though the number of samples to draw as line plots can be modified manually (see below), the statistical analysis is performed with the complete dataset.

Open in a separate window

Fig. 1

Snapshot of Wanderer Web page. Illustrative example of the DNA methylation (left window) and gene expression profiles (right window) associated with MLH1 gene in colon cancer dataset. The left panel in each window includes data selection (gene name/ID and dataset) and view customization parameters (graphics options, statistical threshold, etc.). The right panel of each window displays individual and summarized profiles for normal and tumor samples. The DNA methylation window allows interrogating the correlation between DNA methylation of individual probes with gene expression. At the bottom of the window, direct links provide data download and customized access to other Web portals containing TCGA data

Wanderer complements the DNA methylation profiling with a tool to overview the influence of DNA methylation on gene expression. The user is allowed to navigate through any of the HumanMethylation450 probes located in the displayed genomic region to explore its correlation with the gene expression values.

The generated plots and comprehensive tables containing all the relevant information (individual data, statistical analyses, and additional clinical data) may be downloaded for further data analysis or representation with any software the investigator is used to (e.g., Excel or any other spreadsheet or graphing tool). In addition, the tool allows local navigation and zoom in/out within the region of interest as well as simple graph customization as described below.

Discussion

Integration and mining of TCGA data is a hot topic in clinical bioinformatics. Given the massive nature of data, some of the tools are primarily focused on data retrieval, such as TCGA-Assembler [14], and further annotation and indexing, such as Firehose [15]. A second group of analytical platforms are devoted to provide a comprehensive view of the multiple layers of TCGA by linking molecular data to the clinical features. An outstanding example is cBioPortal [10], which delivers multiple outputs including survival analysis. Regulome Explorer [11] shows associations between clinical and molecular features and provides a circular-based visualization of the data, among others. The Cancer Genome Workbench (CGWB) [16] provides many interfaces to integrate and visualize multiple layers of data. Finally, other tools with a broader scope, as the UCSC Genome Browser, have incorporated TCGA data allowing genomic coordinate-based display [17]. A comprehensive list of cancer data visualization tools may be found at TCGA website [18] and recent reviews [5, 6].

These ambitious initiatives are invaluable contributions which facilitate the access and analysis of the vast amounts of data; but they may be deceiving to wet lab researchers and clinicians with limited computational skills and who are just willing to query very specific information. To face these needs, some tools have been developed providing a very simple interface specialized in a few layers of information. Remarkable examples are MethHC [19] and MENT [20], mainly focused on the comparison of gene expression and DNA methylation.

Wanderer offers a gene-centered access to gene expression and DNA methylation TCGA data. The most outstanding and differential feature of this tool is data display and analysis in a regional framework. The intuitive, simple, and interactive interface facilitates the visualization, interpretation, and obtention of otherwise complex information not easily reachable in other packages. Direct view of the expression and DNA methylation profiles in a regional context may have important applications as illustrated by the usage examples reported above. A quick view of the profile in a specific dataset should be extremely useful to design the most appropriate strategy (i.e., choosing the region or the probe to be analyzed) to determine either the expression of the DNA methylation of a given gene in an experimental setting. These scenarios are common in molecular studies, and the utility of the information is not limited to cancer research. Basic and clinical researchers working in any field of human biology may be also interested in exploring the features of their favorite gene, especially in normal tissues.

Another important feature provided by Wanderer is the simple interface to download small subsets of data in a format easily amenable for customized statistical and graphical analyses. Any user with average proficiency in office suites (Microsoft Office, OpenOffice, IWorks, etc.) or statistics packages (SPSS, SAS, R, etc.) can obtain the desired dataset in a compatible format by just by pressing a button.

Conclusions

Wanderer is an intuitive interface to explore and interpret gene-associated profiles of expression and DNA methylation for all the cancer types available at TCGA. Normal-tumor paired comparisons are readily provided in the form of graphs and comprehensive tables, facilitating the selection of candidate loci to be considered in experimental and statistical settings. The outputs may be downloaded and shared via API. Wanderer is freely accessible at http://maplab.cat/wanderer.

Methods

The user-friendly Web application has been developed using R/shiny and a PostgreSQL backend using an eXtreme programming software development methodology. Wanderer is compatible with most common Web browsers and operative systems, requires no installation, and can be run on commodity computers, such as low-memory laptops.

Data consists of HumanMethylation450 BeadChip and Illumina HiSeq exon and gene quantifications as computed by the RNA-Seq Version 2 pipeline. Data was downloaded with TCGA-Assembler [14]. Source code is available under the GPL v2 terms at [21].

Acknowledgements

Footnotes

Contributor Information

Anna Díez-Villanueva, Email: gro.cppmi@zeida.

Izaskun Mallona, Email: gro.cppmi@anollami.

Miguel A. Peinado, Phone: (34)935543050, Email: gro.cppmi@pam.

References