Background

RNA-sequencing (RNA-Seq) is currently the preferred technology for genome-wide transcriptional profiling due to its combined ease of use, quality of data and suitability for a diverse range of applications [1, 2]. Recent advances in next generation sequencing (NGS) technologies coupled with decreases in the cost of sequencing have resulted in collection of large volumes of RNA-Seq data from many species [3, 4]. These data are typically deposited in online repositories in formats that are text and/or table-based. Visualization of data is a key early step in transcriptomic analysis for many biologists, allowing examination of data quality, as well as rapid interrogation of leads and hypothesis generation. Many researchers who wish to investigate public transcriptome data are not computational experts, for whom transferring data from the format of online repositories to visualization tools is challenging. This creates a barrier to data reuse. The eFP browser, which was first developed for in silico gene expression analysis in Arabidopsis, is an excellent piece of software to display transcriptome data visually [5]. At the time of writing, 20 plant transcriptome data-sets are available publicly in dedicated eFP browsers (http://bar.utoronto.ca, [5,6,7,8,9,10,11,12,13,14,15,16,17]). Degust is a web-based data visualization tool that provides different functionality from eFP functionality (https://github.com/drpowell/degust). It enables users to view all transcripts from all samples in an experiment, examine trends between samples, to visualize quality-control metrics and to drill down into subsets of transcripts with expression patterns of interest. These two data browsers could be integrated to provide users with an easy to use tool for accessing and analysing multiple data-sets and, by developing some enhanced functionality, they could be used for data download and to generating quality images for presentations or publications.

RNA-Seq is often performed at whole plant or organ level using samples that are composed of different tissues and cell types. This approach masks cell- or tissue-specific information about transcripts, which is important to understand spatial-regulation and functions of genes [18, 19]. Spatial resolution is also important to capture transcripts that are expressed at extremely low levels in specific cell types and that are consequently below the limit of detection in bulk samples of tissues [2]. Temporal gene expression data is also an important tool, which can be used to investigate the mechanisms of genome regulation and to understand the relationships between development and gene function [20]. These approaches have been used in functional studies aimed at deciphering regulatory and structural gene networks of diverse plant species, including forest trees and major crops such as wheat (Triticum aestivum), rice (Oryza sativa), maize (Zea mays), barley (Hordeum vulgare), and tomato (Solanum lycopersicum) [21,22,23,24,25,26,27].

Here we present AgriSeqDB (https://expression.latrobe.edu.au/agriseqdb), a web-based resource that can serve as a public portal for accessing, analysing and visualizing tissue and cell-specific transcriptome data-sets from multiple species. Our focus in this implementation is primarily upon transcriptome data-sets during the development of seeds and fruits of agriculturally-relevant species. The database integrates two existing open-source browsers and enhances their functionality. The Degust browser provides access to information on genome-wide expression across samples and data-sets, aiding the discovery of new genes that can contribute to crop improvement. It also provides quality-control information. The eFP browser allows users to visualize between different samples the abundance of individual transcripts encoded by genes of interest.

Construction and content

Database/website architecture

The main structure of AgriSeqDB is described in Fig. 1. It consists of a landing portal that is implemented using an HTML frontend and Python/Django backend to present all data-sets and associated meta-data to users. The landing portal allows the user to discover the data-sets and navigate to data viewers of interest. The existing eFP browser, which has HTML (frontend) and Python (backend) tools, was selected in order to allow users to view expression data on a gene-by-gene basis [5]. Additionally, the existing tool Degust is included to allow viewing of expression profiles across all (or a subset of) genes at once [28]. Degust uses an HTML/Javascript frontend and Haskell backend. Both tools were linked and wrapped within the Landing Portal to ensure that users receive a consistent look and feel when using the portal and each viewer (Fig. 2). The source code for the landing portal and integrations with the viewers is available for reuse (https://bitbucket.org/arobinson/agribiohvc). This repository makes use of git submodules to link the source code of eFP and Degust browsers, each of which was modified slightly from original versions to ensure that they link cleanly; source code for modified versions is available at https://bitbucket.org/arobinson/efp and https://github.com/andrewjrobinson/degust, respectively. The Landing Portal and eFP browser use a MySQL database server to store settings and data/meta-data, while Degust uses files on the file system. A central configuration portal was added to ease the loading of data-sets into the database and of the landing portal documentation, allowing organism annotation upload, data-set upload, data-set configuration such as making it private/public, providing external links and abstract etc., and deploying the data-set to what we refer to as GeneView (eFP) or GeneExplore (Degust).

Fig. 1
figure 1

High level structure of AgriSeqDB showing the linkage between data browsers and the central Landing Portal. The Landing Portal provides a central place to access all data-sets and provide meta-data that isn’t provided by data browsers. The data browsers provide access to the same data in various forms to enable greater insight

Full size image
Fig. 2
figure 2

AgriSeqDB home screen showing the six data-sets from species including crops species of major agricultural importance that are currently in the database

Full size image

Data sources

All data-sets displayed currently in AgriSeqDB are transcriptomes published recently and deposited in public databases (Fig. 2, Table 1). Users of AgriSeqDB can view data directly from database server without the need to download it and then install/configure a viewer to visualise it. The data-sets were generated in six studies of seeds or fruit. The first is a study we conducted of transcriptome changes in whole Arabidopsis seeds during germination, which provides a useful reference due to this species’ high-quality genome sequence and annotation [20]. Additionally, we displayed five data-sets from major agricultural crops. These were: A study of transcriptome changes in different tissues (aleurone, starchy endosperm, embryo, scutellum, pericarp, testa, husk and crushed cell layers) of barley grain at different stages of germination [21]; a study on transcriptome changes associated with different cell types of maize endosperm after pollination [22]; a study on transcriptome changes associated with seed germination and coleoptile growth in rice [23]; a study on transcriptome changes associated with fruit development in tomato fruit [26]; and a study on grain/endosperm transcriptome of bread wheat [24]. The GeneExplore (Degust) component of AgriSeqDB requires RNA-seq data in raw count format (i.e. number of reads per gene or transcript, not normalised) for the subsequent analyses it applies. We made use of the raw count data provided by the original authors on the respective GEO/SRA repositories for the Arabidopsis, wheat, rice and barley studies (Table 1). In those cases, mapping and read counting were consequently as described previously [20, 21, 23]. In the case of the maize and tomato data-sets, raw count data were not available from the GEO/SRA repositories, but the original sequence reads were. For these we aligned and quantified the count data using Kallisto with the reference tomato or maize transcriptomes (Solanum lycopersicum SL2.50 or Zea mays AGPv4), using the resulting data as the input to AgriSeqDB [29].

Table 1 RNA-Seq data-sets included in AgriSeqDB
Full size table

Utility and discussion

Our goal was to develop a publicly accessible transcriptome database that provides simple and readily available tools to perform functional analysis of individual target genes or sets of genes. AgriSeqDB is a highly interactive and multi-view database that can be used for various purposes, including the discovery of genes of interest. Users of AgriSeqDB can view data directly from database server without the need to download it and then install/configure a viewer to visualise it. However, we provide the option for advanced users to download and install their own local AgriSeqDB for custom data-sets.

GeneView (eFP)

AgriSeqDB also allows users to get a better understanding of individual genes of interest, by inspecting them within GeneView (eFP) (Fig. 3). This incorporates the full existing functionality of eFP [5]. Users can visualise expression of transcripts across all samples so that they may consider the relationships between samples (i.e. growth stage, tissue type, various treatments). Additionally, we incorporated an additional image download function, not previously available. Images may be downloaded in high-resolution .png format for presentations or publications. This is done by single clicking the Download button (Fig. 3). We have also enabled cross-species comparisons directly from the GeneView (eFP) records. When users are viewing a gene that interests them within GeneView (eFP), they can click on a button that directly returns a search from the Gramene database (http://www.gramene.org). This returns homologs, orthologs and paralogs drawn from 2,076,020 genes across 53 crop and model plant species, as well as a comparative phylogenetic tree.

Fig. 3
figure 3

The full screenshot showing AT2G40170 gene expression in GeneView (eFP) browser. The user uses the search form at the top to select the gene of interest and select the mode of operation including: (1) absolute, shows the counts as stored in the database for the primary gene, (2) relative, shows the counts relative to the control for primary gene, and (3) compare, counts as a ratio between the primary and secondary genes. Clicking the view button updates the figure below to show the expression levels of each sample by colour coding the fill area with a scale red-yellow (for absolute) and red-grey-blue (for relative and compare). Alternatively, the user can click the download button (indicated by a green arrowhead) to download the expression image at twice the resolution as shown on-screen (ready for publication). Data is from transcriptome of Arabidopsis seeds during germination [20]

Full size image

GeneExplore (Degust)

Users are presented with a simple interface to query all genes using GeneExplore (Degust) (Fig. 4a). Extensive existing functionality is available to users within Degust [28]. Filters can be created on the data based upon expression levels in individual samples, false discovery rate (FDR) and Log2 fold-change cut-off. Sub-sets of samples or transcripts can be selected for analysis can be analysed and the sample for referencing fold-change can be selected. MA plots of comparisons between pairs of samples can be displayed (Fig. 4b). Data quality metrics can be assessed by inspecting whether the replicates of each sample group together in the multidimensional scaling (MDS) plot (Fig. 4c). Data tables can also be downloaded for selected transcripts in .csv format for downstream analyses.

Fig. 4
figure 4

Screen shot of the GeneExplore (Degust) browser and subsequent result pages. a the Arabidopsis time-series data-set is shown here as an example, displaying transcripts that are up-regulated in S samples but down-regulated in SL samples (Top panel). The user can select which samples they wish to see with the checkboxes in the top left of screen along with the method of analysis (voom/limma, edgeR, or voom). In the top right, the user can control the rendering and thresholds of using the options dialog. All genes that match filters above are shown in a heat-map, which clusters genes with similar levels of expression (Middle panel). Running the mouse-over each gene highlights it in the plots above. Table showing all matching genes in tabular format with the expression levels for each sample, false discovery rate and any extra annotation columns provided in the data-set (Lower panel). In the top centre the user can limit genes by using 1 of 3 interactive plots, and the parallel coordinates plot allows the user to limit genes by their log fold gene expression (per sample). b Example of an MA plot. Users can limit genes by drawing a box around genes on the on the MA plot; the two samples used for the MA plot are specified in the Options dialog (top right). c An MDS plot showing groupings of the individual replicates of each sample. Data is from transcriptome of Arabidopsis seed during germination [20]

Full size image

Data-set administration (advanced usage case)

One key inclusion to AgriSeqDB is the data-set administration tool. This tool is available only when users download and install their own local AgriSeqDB, for reasons of server security detailed below. eFP browser did not contain an interface to upload data, so configuration required much manual interaction. While Degust contained its own administration tool, it was not flexible enough to accommodate eFP and the landing portal. Consequently, we developed a new data upload interface to encompass both eFP and Degust. This allows the user (secured by username/password) to upload new data-sets and deploy them to each of the viewers. The tool provides the user the ability to execute custom code on the host server, access to which should be restricted to the local database administrator and trusted users. An example configuration is included in Fig. 5 and Additional file 1: Figure S1. A link is provided on our AgriSeqDB landing portal that takes users to a repository from which all AgriSeqDB code can be downloaded and from where installation/administration instructions and help files can be viewed (Fig. 2). The direct address of the code and help repository is https://bitbucket.org/arobinson/agribiohvc. The repository includes a link from which users can access a Galaxy Project RNA-seq analysis tutorial (https://galaxyproject.org/tutorials/rb_rnaseq/), which users may find useful to prepare data when establishing their own local AgriSeqDB.

Fig. 5
figure 5

The process of uploading a data-set to a local installation of AgriSeqDB consists of 3-steps. First, the user selects a unique identifier and display name for the data-set (a). Second, the user chooses the count file and various eFP images used to display expression values from their PC to upload (b). Finally, the user can alter many settings that control how the data-set is displayed in the landing portal and each data-viewer as shown in Additional file 1 (see Supplementary Fig. 1). Most settings have sensible defaults where user input is required, time saving tools (such as colour picker & clickable image for eFP settings) or spreadsheet import/export (sample settings)

Full size image

Conclusions

We believe AgriSeqDB will be an important resource and data-reuse tool for plant biologists who seek greater insights into the role of individual genes or group of genes in biological processes, including for comparative studies in crop species of major agricultural importance. The databases will be periodically updated with more viewers and data-sets, focusing on additional tissue and cell-specific data-sets from crop species. The database currently contains results of RNA-Seq from different tissues and cell types, and it is planned that transcriptome data from single cell RNA-Seq will be added in the future. In the long term it is envisaged that users will be provided with links to GEO auto-download and view as well as allowed to upload data-sets at least temporarily. All source code is freely available for reuse by advanced users.