The mouse Gene Expression Database (GXD): 2021 update (2024)

Comprehensive literature survey

We systematically survey journals to find all publications examining endogenous gene expression during mouse development. In a first curation step for each paper, we annotate the genes and ages analyzed and the expression assay types used. Annotations are based on the entire publication, including Supplemental Data, and employ official nomenclature for genes. This information, combined with bibliographic information from PubMed, is accessible via the Gene Expression Literature Search form (http://www.informatics.jax.org/gxdlit). GXD’s literature content records are comprehensive and up-to-date from 1990 to the present. GXD has records for >28 400 references and >16 200 genes. The Gene Expression Literature Search provides scientists with an effective tool for finding publications with specific expression data, and it helps GXD Curators to prioritize papers for detailed expression annotation.

Detailed expression data

In a second curation step, the expression data are annotated in detail. For each expression assay, we record information about the gene studied, the strength and pattern of expression in specific anatomical structures, the probes and experimental conditions used, and the age and genetic background of the specimen(s) analyzed. Images of the data accompany the annotations when available. Standard gene, mouse strain and allele nomenclature, controlled vocabularies and an extensive anatomy ontology are employed to enable thorough data integration and search capabilities. See http://www.informatics.jax.org/assay/MGI:2673718 for an example of an RNA in situ record, and Figure ​Figure4C4C for an example of immunohistochemistry records (partial records shown). As of September 2020, GXD contains detailed expression data for nearly 15 000 genes, including data from numerous strains of wild-type mice and from >4900 mouse mutants. GXD now holds >365 000 images and >1.74 million expression result annotations for classical types of expression data. The majority of these data (85%) are from in situ hybridization, immunohistochemistry and in situ reporter (knock in) experiments.

RNA-Seq and expression microarray metadata index

GXD now provides a searchable metadata index of mouse high-throughput expression experiments available in public repositories (11). We incorporate mouse RNA-Seq and expression microarray experimental metadata from the Gene Expression Omnibus (GEO) (12) and ArrayExpress (13), and apply GXD annotation standards to samples and attributes of experiments that meet GXD’s scope of endogenous expression. This includes experiments that examine endogenous expression in wild type and mutant mice, and cover the entire life span of the laboratory mouse (all pre- and postnatal stages). Researchers can now find comprehensive sets of high-throughput experiments using standardized search terms from controlled vocabularies and ontologies. This task is not possible when conducting searches through repository resources, since standardized metadata are not required for data submission, and term heterogeneity is widespread.

We load mouse experiment-level metadata (ID, title, abstract and experiment type) from ArrayExpress weekly (Figure ​(Figure1A).1A). A dedicated curation tool facilitates efficient experiment evaluation and sample and experiment attribute annotation to standardized terms (not shown). An initial triage step identifies experiments that are considered consistent with GXD scope. Manual evaluation at this step is supported by supervised machine learning (a linear SVM classifier, see (11) for more detail) which has proven to be effective for the classification of out-of-scope experiments. Sample metadata for GXD-relevant experiments are downloaded on demand into the curation tool, annotated to standard terms and added to the index. Of 16 782 high-throughput expression experiments downloaded from ArrayExpress, 3163 were considered relevant to GXD’s scope (2463 microarray, 700 RNA-Seq). At the time this load was developed, ArrayExpress included all experiments from GEO. This is no longer the case. GXD plans to extend this load to acquire data directly from both ArrayExpress and GEO, assuring complete representation from both repositories.

RNA-Seq expression data in GXD

GXD has incorporated high quality RNA-Seq expression data for GXD-relevant RNA-Seq experiments in GXD’s RNA-Seq and Expression Microarray Metadata Index. These data are imported from the Expression Atlas at EBI (14), and then processed further by GXD to allow full integration into our system, and accessibility through our search and display tools. The Expression Atlas incorporates primary data for high-quality RNA-Seq data sets from ArrayExpress and GEO, and applies a processing pipeline designed to leverage state-of-the-art annotation methods to generate reliable, updated TPM values for genes represented in current Ensembl releases (14). We download these TPM values, and prepare the data for integrated access using GXD’s RNA-Seq processing pipeline (Figure ​(Figure1B).1B). Expression data for new GXD-relevant RNA-Seq experiments from the Expression Atlas are incorporated into GXD as they become available.

A primary goal of GXD’s RNA-Seq processing pipeline is to identify the unique biological replicate sets for each experiment; determine the averaged quantile normalized TPM value for each gene per biological replicate set; and assign each of these TPM values a Present/Absent call in GXD. RNA-Seq expression data files from the Expression Atlas provide TPM values per gene for each technical replicate (run) of a given experiment (TPM/gene/runID/experiment). The samples from which technical replicates are derived (often multiple runs per sample) are not specified in the TPM data files. The GXD metadata index, however, is annotated at the sample level (technical replicate identifiers are not included in the index). Thus, to join GXD-curated sample metadata with Expression Atlas TPM values, a second load of run IDs per sample for each experiment is necessary from ArrayExpress (see Figure ​Figure1B).1B). TPM values per gene from technical replicates are averaged to account for any technical variation. Then, to lessen the influence of biological sampling variation, and to avoid excessive redundancy in GXD user interface displays, sample-level TPM values for biological replicates are combined. Since biological replicate information is not provided in the files downloaded from the Expression Atlas or ArrayExpress, we take advantage of GXD-curated metadata to identify bioreplicate samples for each experiment. Samples from the same experiment that share all metadata field values are considered bioreplicates. Bioreplicate sample TPM values for each gene are consolidated in a two-step process. For each experiment, sample-level TPM values for bioreplicate samples are first quantile normalized (15) using python numpy in the Pandas DataFrame, and then these quantile normalized (QN) TPM values are averaged for each gene across the bioreplicate sample set. This results in a single TPM value per gene per set of sample bioreplicates. We store bioreplicate set information in the database and assign each bioreplicate set a unique identifier (see Figure ​Figure4A).4A). Finally, we map averaged QN TPM values to the same TPM range bins used by the Expression Atlas (see Figure ​Figure1B),1B), and assign a Present/Absent value based on TPM values relative to the Below Cutoff value range (<0.5 TPM = Absent expression in GXD). GXD uses the same Present/Absent values for classical expression assays, thus this last step allows us to fully integrate RNA-Seq data with classical expression data in GXD searches, filters and displays (see Figure ​Figure2).2). To date, we have loaded expression data for 70 RNA-Seq experiments in the GXD metadata index. These include 1846 distinct samples that were condensed to 631 biological replicate sets, representing 88 distinct anatomical structures and 85 distinct mouse strains. Mouse genes represented in Expression Atlas RNA-Seq TPM files cover the entire Ensemble transcriptome (nearly 55 000 genome features), including protein-coding and non-coding RNA genes. Total RNA-Seq assay results amount to nearly 35 million, with comprehensive genome coverage for each experiment.

RNA-Seq and microarray experiment search

The RNA-Seq and Microarray Experiment Search tool [http://www.informatics.jax.org/gxd/htexp_index; detailed in (11)] enables users to quickly and reliably find specific high-throughput expression data sets from the metadata index described above. The search form supports detailed queries for RNA-Seq and/or transcription microarray experiments by attributes of the samples assayed, including anatomical structure, developmental stage, mutant gene, mouse strain and sex. The search form also provides a text search function that returns experiments by text string matches in titles or descriptions. In addition, users can search for specific experiments by experiment ID.

The search results summary provides well organized information profiles of each experiment returned, which include title and full description, assay type (method), study type and experimental variables involved (i.e.the metadata fields that distinguish samples in the study). Access to GXD-curated metadata details for each sample of an experiment is prominent on the experiment summary. The total number of samples is shown and the number of samples that matched the search parameters is also shown. This is a valuable feature, since experiments can include samples from diverse biological sources. A link is provided to a tabular view of all samples for an experiment with their associated metadata, and the samples that matched search parameters are distinguished. When experiments are returned by text string match, matching text is conveniently highlighted. The summary also includes links to relevant external resources for each experiment where available, including PubMed, ArrayExpress, GEO and the Expression Atlas. Finally, for RNA-Seq experiments included in GXD’s RNA-Seq expression data load, links are provided to the GXD expression summary page for the corresponding experiment.

New and integrated search capabilities for RNA-Seq and classical expression data

Our approach is to add value to RNA-Seq data by integration with the other expression data types in GXD, and with the genetic, functional, phenotypic and disease-oriented data in MGI. This integration allows us to provide new search and analysis capabilities for RNA-Seq data. The newly incorporated RNA-Seq data are accessible throughout much of the GXD user interface, including (a) the Standard Gene Expression Data Search, which permits querying for expression data by an array of different search parameters, including user-specified metadata and biomedically relevant annotation profiles (http://www.informatics.jax.org/gxd; Standard Search tab); (b) the GXD Batch Search tool, which returns expression data from input lists of genes (http://www.informatics.jax.org/gxd/batchSearch) and (c) the Mouse Developmental Anatomy Browser, where links to expression data from specific anatomical structure searches are superimposed on a dynamic hierarchical display of mouse anatomical structures (http://www.informatics.jax.org/vocab/gxd/anatomy/). RNA-Seq data are excluded from the default settings of the Standard Gene Expression Data Search due to the expansive result sets common to genome-wide assays; this is clearly indicated on the Search Form. RNA-Seq data can be included with a single click in the Assay types section on the form. To assure responsive system performance, we optimized front end configuration and imposed reasonable limits for Standard and Batch GXD searches. For the Standard Gene Expression Data search, we set a limit of 21 million expression assay results. This threshold is high enough to include all results for any anatomical structure or developmental stage spanning all assay types (including RNA-Seq). The interface informs users when their search results pass this threshold, and encourages search refinement. The GXD Batch Search has an input limit of 5000 genes.

All of GXD’s expression data search forms, and links provided from the Gene Expression section of MGI Gene Detail pages lead to the same multi-tabbed displays that summarize data at different levels of detail (Genes, Assays, Assay results, Images) and via two different Matrix Views (Tissue ×Stage and Tissue ×Gene). RNA-Seq data are integrated into these summary views when selected on the Standard search form. Tools provided on these summaries allow users to filter, sort, and iteratively refine search results, and download the data for further analysis. We have recently added new gene-level filters that enable users to narrow search results effectively (see Figure ​Figure2B2B).

Figure ​Figure22 shows results of a search that would be of interest to researchers who study macrophage function in the liver. Search parameter details are displayed in the upper left of the summary (Figure ​(Figure2A),2A), and users have the option to open the query form and modify their search at any time. A suite of filters that allows refinement of expression results is shown in Figure ​Figure2B.2B. A new quantitative expression filter that restricts RNA-Seq data sets by TPM Level was added to pre-existing expression result filters (left-hand column). Six new gene-level filters (right-hand column) have also been added that greatly expand analytical capacity, as users can restrict expression results to sets of genes by gene type and by annotations to high-level terms of the Gene Ontology (GO, 16), Mammalian Phenotype Ontology (MP, 17) or Disease Ontology (DO, 18) (Figure ​(Figure2B).2B). Separate filters are combined by Boolean AND, while multiple choices within a filter are combined by Boolean OR. Expression results from liver were refined by applying a set of gene-level filters designed to target genes that have GO and MP annotations consistent with macrophage function (see Filtered by list in Figure ​Figure2B).2B). These filters reduced the initial assay result set by over 100-fold, by dropping the genes returned from 54 972 (not shown) to 504.

Figure ​Figure2C2C features assay results (partial) for the filtered search in Figure ​Figure2B,2B, showing integrated expression data from RNA-Seq and Immunohistochemistry assays. Results for two of the 504 genes are shown (Adgrb1 and Adgre1). New metadata columns were added to the Assay Results tab to distinguish bioreplicate sets (toggled open in Figure ​Figure2C).2C). RNA-Seq data-specific columns include TPM Level, the number of biological replicate samples and a Notes field used to record salient metadata features not accounted for under other metadata fields. Each RNA-Seq data row in the table represents the expression record (with averaged QN TPM value mapped to a TPM Level) from that set of combined bioreplicate samples for the corresponding experiment. RNA-Seq rows also feature novel links to the Expression Atlas, which offers additional display and analysis tools, and to the GXD RNA-Seq and Microarray Experiment Summary for the corresponding experiment. All RNA-Seq results in the table can be viewed in a heat map rendered by the Morpheus resource (https://software.broadinstitute.org/morpheus/) (discussed below), and all expression results from the user's search can be exported to a text file for further analysis elsewhere. To illustrate the benefits of integrated expression results from RNA-Seq and classical expression assays, a link to GXD assay details in the Images column for one of the immunohistochemistry assays is indicated (detailed in Figure ​Figure4C).4C). The other summary views (Genes, Assays, Images, and the Tissue ×Stage and Tissue ×Gene Matrix tabs) are described elsewhere (8). Slight modifications were made to the Assays tab and the Tissue x Gene Matrix to accommodate the whole genome perspective of RNA-Seq data. For the Assays tab, instead of listing all genes for an assay (as done for classical assay types), each RNA-Seq experiment returned is displayed in a single row, with the option to filter the expression result set by that experiment (Figure ​(Figure3).3). Links are also provided to the RNA-Seq and Microarray Experiment summary page for each RNA-Seq experiment. For the Tissue ×Gene Matrix, column pagination (genes) was introduced to manage display of the vast number of genes that can be included in RNA-Seq results (not shown).

Morpheus heat map of GXD RNA-Seq data

We have embedded the Morpheus heat map visualization and analysis tool from the Broad Institute (https://software.broadinstitute.org/morpheus/) into the GXD interface. When RNA-Seq results are present, the expression results summary page provides a link to a Morpheus heatmap (see Figure ​Figure2C).2C). This gives users one-click access to the wide range of display and analysis tools offered by the Morpheus resource for RNA-Seq data sets tailored by GXD’s powerful search and filtering functions. The Morpheus heat map rendered for the RNA-Seq results from the GXD query in Figure ​Figure22 is shown in Figure ​Figure4.4. The heat map displays quantitative expression (using a color-coded average QN TPM value range) for each gene (rows) and the corresponding biological replicate sample sets (columns) returned from the GXD query (Figure ​(Figure4A4A).

Columns in the heat map grid represent biological replicate sets. Column labels are derived from the anatomical structure, experiment ID and GXD-assigned bioreplicate set ID of featured biological replicate sets. Stars shown in column labels indicate that the samples were derived from mutant mice, which is important when interpreting expression profiles. Wild type samples have no star in the column label. Morpheus provides column data display tools to help users recognize patterns in the data, including colored metadata rows just below the column labels. The tool assigns separate colors to distinct metadata annotations in each metadata row making it easy to spot common annotations across the sets of bioreplicates (underlying values are displayed on mouse-over). By default, Morpheus suppresses display of metadata rows that are shared by all columns, but users can include these common rows if desired. Morpheus offers primary and secondary metadata sorting by simply clicking on metadata row labels, while additional sorting and filtering options are available (for columns and rows) via the main display tools (see Figure ​Figure4B4B).

Rows in the heat map grid represent genes. The default label for gene rows is Gene Symbol, MGI ID and Ensembl gene ID (reduced to Gene Symbol in Figure ​Figure4A).4A). Default TPM value color code ranges (shown in the heat map legend in Figure ​Figure4B)4B) were designed to approximate display settings used by the Expression Atlas, facilitating comparison between resources, and to manage the enormous dynamic range of RNA-Seq data (color saturation set at 5000 TPM). Morpheus display options allow users to adjust the colors and TPM ranges of the heat map, to improve resolution in any TPM value range desired. The TPM value for any cell is displayed by mouse-over. It is possible to generate a heat map that has some empty cells; these are colored white and display ‘NaN’ (Not a Number) on mouse-over. This can happen when an expression value filter (TPM Level or Detected?) was applied to the GXD result set before rendering the heat map (since genes can have a range of TPM values across the bioreplicate sets present). This can also happen when an experiment has fewer total gene annotations (if not updated to the latest Ensembl build, for example). The ability to align metadata values with TPM value ranges allows users to spot biologically significant expression profiles, such as sex- or strain-specific effects and expression effects of specific mutants compared to wild type samples. Morpheus also offers a range of data clustering tools under the Tools dropdown menu (not shown). To keep response times manageable, we’ve imposed a 10 million result limit for export to the heat map. A processing progress bar is displayed while the heat map is generated.

An example of the value of integrated RNA-Seq and classical expression assays is shown for genes Adgre1 and Clec4f in Figure ​Figure4A4A andC (marked with carets for comparison). GXD assay results of immunohistochemical staining in mouse liver reveal spatially-restricted coexpression of these two genes in Kupffer cells (which are resident liver macrophages), while the RNA-Seq heat map provides a quantitative expression survey across the bioreplicate samples included.

1. Bult C.J., Blake J.A., Smith C.L., Kadin J.A., Richardson J.E. the Mouse Genome Database Group . Mouse Genome Database (MGD) 2019. Nucleic Acids Res. 2019; 47:D801–D806. [PMC free article] [PubMed] [Google Scholar]

2. Smith C.M., Hayamizu T.F., Finger J.H., Bello S.M., McCright I.J., Xu J., Baldarelli R.M., Beal J.S., Campbell J.W., Corbani L.E. et al... The mouse Gene Expression Database (GXD): 2019 update. Nucleic Acids Res. 2018; 47:D774–D779. [PMC free article] [PubMed] [Google Scholar]

3. Drabkin H.J., Christie K.R., Dolan M.E., Hill D.P., Ni L., Sitnikov D., Blake J.A.. Application of comparative biology in GO functional annotation: the mouse model. Mamm. Genome. 2015; 9:574–583. [PMC free article] [PubMed] [Google Scholar]

4. Ruzicka L., Howe D.G., Ramachandran S., Toro S., VanSlyke C.E., Bradford Y.M., Eagle A., Fashena D., Frazer K., Kalita P. et al... The Zebrafish Information Network: new support for non-coding genes, richer Gene Ontology annotations and the Alliance of Genome Resources. Nucleic Acids Res. 2019; 47:D867–D873. [PMC free article] [PubMed] [Google Scholar]

5. Fortriede J.D., Pells T.J., Chu S., Chaturvedi P., Wang D., Fisher M.E., James-Zorn C., Wang Y., Nenni M.J, Burns K.A. et al... Xenbase: deep integration of GEO & SRA RNA-seq and ChIP-seq data in a model organism database. Nucleic Acids Res. 2020; 48:D776–D782. [PMC free article] [PubMed] [Google Scholar]

6. Antin P.B., Yatskievych T.A., Davey S., Darnell D.K.. GEISHA: an evolving gene expression resource for the chicken embryo. Nucleic Acids Res. 2014; 42:D933–D937. [PMC free article] [PubMed] [Google Scholar]

7. Smith C.M., Finger J.H., Kadin J.A., Richardson J.E., Ringwald M.. The gene expression database for mouse development (GXD): putting developmental expression information at your fingertips. Dev. Dyn. 2014; 243:1176–1186. [PMC free article] [PubMed] [Google Scholar]

8. Finger J.H., Smith C.M., Hayamizu T.F., McCright I.J., Xu J., Eppig J.T., Kadin J.A., Richardson J.E., Ringwald M.. The mouse gene expression database: New features and how to use them effectively. Genesis. 2015; 53:510–522. [PMC free article] [PubMed] [Google Scholar]

9. Hayamizu T.F., Baldock R.A., Ringwald M.. Mouse anatomy ontologies: enhancements and tools for exploring and integrating biomedical data. Mamm. Genome. 2015; 26:422–430. [PMC free article] [PubMed] [Google Scholar]

10. Finger J.H., Smith C.M., Hayamizu T.F., McCright I.J., Xu J., Law M., Shaw D.R., Baldarelli R.M., Beal J.S., Blodgett O. et al... The mouse Gene Expression Database (GXD): 2017 update. Nucleic Acids Res. 2017; 45:D730–D736. [PMC free article] [PubMed] [Google Scholar]

11. Smith C.M., Kadin J.A., Baldarelli R.M., Beal J.S., Blodgett O., Giannatto S.C., Richardson J.E., Ringwald M.. GXD’s RNA-Seq and Microarray Experiment Search: using curated metadata to reliably find mouse expression studies of interest. Database (Oxford). 2020; 2020:baaa002. [PMC free article] [PubMed] [Google Scholar]

12. Clough E., Barrett T.. The gene expression omnibus database. Methods Mol. Biol. 2016; 1418:93–110. [PMC free article] [PubMed] [Google Scholar]

13. Athar A., Füllgrabe A., George N., Iqbal H., Huerta L., Ali A., Snow C., Fonseca N.A., Petryszak R., Papatheodorou I. et al... ArrayExpress update - from bulk to single-cell expression data. Nucleic Acids Res. 2019; 47:D711–D715. [PMC free article] [PubMed] [Google Scholar]

14. Papatheodorou I., Fonseca N.A., Keays M., Tang Y.A., Barrera E., Bazant W., Burke M., Füllgrabe A., Fuentes A.M., George N. et al... Expression Atlas: gene and protein expression across multiple studies and organisms. Nucleic Acids Res. 2018; 46:D246–D251. [PMC free article] [PubMed] [Google Scholar]

15. Hicks C., Irizarry R.A.. When to use quantile normalization?. 2014; bioRxiv doi:04 December 2014, preprint: not peer reviewed 10.1101/012203. [CrossRef]

16. The Gene Ontology ConsortiumThe Gene Ontology Resource: 20 years and still GOing strong. Nucleic Acids Res. 2019; 47:D330–D338. [PMC free article] [PubMed] [Google Scholar]

17. Smith C.L., Eppig J.T.. The Mammalian Phenotype Ontology as a unifying standard for experimental and high-throughput phenotyping data. Mamm. Genome. 2012; 23:653–668. [PMC free article] [PubMed] [Google Scholar]

18. Bello S.M., Shimoyama M., Mitraka E., Laulederkind S.J.F., Smith C.L., Eppig J.T., Schriml L.M.. Disease Ontology: improving and unifying disease annotations across species. Dis. Model. Mech. 2018; 11:dmm032839. [PMC free article] [PubMed] [Google Scholar]