MOET: a web-based gene set enrichment tool at the Rat Genome Database for multiontology and multispecies analyses

Abstract

Introduction

In recent years, functional genomics and transcriptomic data have increased exponentially due to the development of high-throughput technologies, next-generation sequencing in particular (Ghandikota et al. 2018; Hinderer et al. 2019; Raudvere et al. 2019). Thus, database resources are essential to make this information accessible, as are algorithms that facilitate interpretation of large datasets for generating novel hypotheses for functional validation. The Gene Ontology (GO) Consortium was developed as a resource to describe and curate functional similarities for such data to make meaningful interpretations and to computationally represent the current knowledge (Ashburner et al. 2000; The Gene Ontology Consortium 2019, 2021). GO creates a standardized vocabulary to define the biological processes, cellular components, and molecular functions associated with the gene. Currently, GO provides over 7 million annotations across multiple organisms (Boyle et al. 2004; Eden et al. 2009; The Gene Ontology Consortium 2021). These GO terms can be used for gene set enrichment analysis, which is a widely used process to statistically identify terms that are significantly overrepresented or enriched within a list of input genes or proteins (Pomaznoy et al. 2018; Zuniga-Leon et al. 2018). This has led to the development of many web-based applications and programs providing easy access for researchers from diverse scientific disciplines for a variety of uses (Khatri et al. 2004; The Gene Ontology Consortium 2017).

In the past years, numerous ontology analysis tools (Supplementary Table S1) have been developed (Berriz et al. 2003; Subramanian et al. 2005; Mi et al. 2021). Many of these tools are web-based, public, and free to use; others require licensing (e.g. IPA) (Kramer et al. 2014) or downloading a program (program-based) for their use (e.g. GSEA, BinGO) (Maere et al. 2005; Subramanian et al. 2005). A few ontology analysis tools require knowledge of specific programming languages, such as R (e.g. DOSE, GSEA) (Maere et al. 2005; Yu et al. 2015) or Python (e.g. GOATools) (Klopfenstein et al. 2018), or have operating system-specific requirements (e.g. FunRich), making them inconvenient for some users (Benito-Martin and Peinado 2015; Fonseka et al. 2021). Therefore, web-based tools, such as AmiGO, FunSet, and PANTHER are comparatively popular because of their ease to use (Carbon et al. 2009; Hale et al. 2019; Mi et al. 2021). However, most gene enrichment or overrepresentation tools are limited only to the analysis of specific species, and/or ontologies [e.g. WormBase enrichment analysis, Comparative GO (Fruzangohar et al. 2013; Angeles-Albores et al. 2018; Le 2020)].

Most of the gene enrichment or overrepresentation tools adopt a common strategy of entering a set of genes which are then compared statistically against a given background gene set (which may or may not be defined by the user) (Huang da et al. 2009a). Some of the tools display the results of the analysis in directed acyclic graphs (DAG), such as WEGO, GO:Term Finder and WebGestalt (Boyle et al. 2004; Ye et al. 2018; Liao et al. 2019). Of note, the development of ontologies is an ongoing process with terms being created, obsoleted, or merged regularly. Over the last 2 years, within GO, the number of terms created, obsoleted, and merged were 672, 661, and 752 terms, respectively (http://geneontology.org/stats.html; The Gene Ontology Consortium 2021). Several tools that determine ontology term enrichment are limited by a lack of frequent updates resulting in inaccurate outputs; for example, DAVID, FuncAssociate and WebGestalt were last updated in 2016, 2018, and 2019, respectively (Berriz et al. 2003; Jiao et al. 2012; Liao et al. 2019). Also, a few tools do not provide the user with multiple testing corrections for their analysis (e.g. Algal Functional Annotation Tool, Comparative GO) which is an important parameter in functional ontological evaluations due to the large volume of data (Lopez et al. 2011; Fruzangohar et al. 2013).

The Rat Genome Database (RGD) was developed in 1999 as a one-stop rat genomic and physiologic data repository. RGD has progressed to store information for mouse, human, chinchilla, bonobo, dog, pig, naked mole-rat, vervet (or green monkey) and 13-lined ground squirrel along with advanced RGD tools (Laulederkind et al. 2019; Smith et al. 2020). A major strength of RGD is its expert manual curation of genes demonstrated by 251,278 cumulative manual annotations. The analysis and visualization of all these data are facilitated by RGD tools such as advanced search tool OLGA (Object List Generator and Analyzer), InterViewer (protein-protein interaction viewer), and GOLF (Gene-Ortholog Location Finder) (Smith et al. 2020), with each of them providing unique features. In addition to manual curation efforts, RGD also imports data from other databases to provide the user with further information for their analysis. To improve on gaps in the term enrichment field and to meet the growing complex experimental needs of the research community, RGD has developed the Multi Ontology Enrichment Tool (Supplementary Table S1) (MOET, https://rgd.mcw.edu/rgdweb/enrichment/start.html). MOET is a unique web-based ontology analysis tool that generates a list of terms statistically overrepresented within the user's genes of interest, leveraging multiple classes of ontology annotations. Some of the advantages of MOET over the currently available enrichment tools are:

• It is a web-based, publicly available, and freely accessible ontology analysis and visualization tool.
• It is simple to use, and results are provided with a few clicks in seconds; no software installations or programming skills are required.
• It provides functionality for multiple ontologies, including Disease, GO, Pathway, Phenotype, and Chemical entities (ChEBI).
• It provides enrichment analysis for multiple RGD species, including rat, mouse, human, bonobo, squirrel, dog, pig, chinchilla, naked mole-rat and vervet (green monkey).
• The P-values are displayed for each term in the output with Bonferroni multiple testing corrections to control false positives.
• It supports input of any of 11 different common identifier types, saving the user from translating one type of ID to an acceptable input ID.
• It is updated weekly, providing the user with the most recent data for analyses.

Methods

Data that support the tool

The backend database for MOET consists of RGD's extensive corpus of functional annotations derived from manual curation at RGD, supplemented with automated pipelines that import and integrate data from multiple databases (Fig. 1). The curators at RGD use in-house designed curation software integrated with an RGD-developed literature search tool OntoMate (Liu et al. 2015) to identify peer-reviewed journal articles related to a specific disease category and create annotations based on genes for RGD species, in addition to annotations imported from other data sources. Disease, pathway, and ChEBI annotations for rat, mouse, and human studies are annotated at RGD with appropriate evidence codes for the data types. Annotations are assigned to orthologous genes in other species using the inferred from sequence orthology (ISO) evidence code. Rat-specific GO annotations and rat and human gene-specific phenotype annotations are curated manually from the same gene list at RGD. To provide integrated disease-focused environments, RGD has developed 15 Disease Portals (Shimoyama et al. 2016) (as of September 2021); the portals are designed to be entry points for disease-focused researchers to access integrated information related to their area of interest including disease-specific annotations, tools and datasets.

Open in a separate window

Fig. 1.

Schematic overview of MOET functionality. The user provides input using the options shown, MOET uses data integrated at RGD from all the sources, performs enrichment and statistical analysis, and provides downloadable results.

The data sources for imported ontology annotations for rat, mouse, and human include: UniProt’s Gene Ontology Annotation group (UniProt-GOA) for rat GO annotations from other groups, including rat Inferred from Electronic Annotation (IEA) annotations (UniProt Consortium 2021); Gene Ontology Resource for human and mouse GO annotations (The Gene Ontology Consortium 2021); Mouse Genome Informatics for Mammalian Phenotype (MP) Ontology annotations for mouse, and disease annotations for human and mouse (Baldarelli et al. 2021); Online Mendelian Inheritance in Man (OMIM) for human disease data (Amberger and Hamosh 2017); Online Mendelian Inheritance in Animals (OMIA) for disease and phenotype annotations for dog and pig (Nicholas 2021); ClinVar for human disease annotations (Landrum et al. 2018); Comparative Toxicogenomics Database (CTD) for rat, mouse and human gene-chemical interaction annotations and human disease annotations (Davis et al. 2019); Human phenotype ontology group for HPO annotations (Köhler et al. 2021); and Small Molecule Pathway Database (SMPDB) for human pathway annotations (Jewison et al. 2014). In addition to the regular import of current data, RGD has data archived from the Kyoto encyclopedia of genes and genomes (KEGG) for pathway annotations (Kanehisa et al. 2021) 8 years ago. Also, RGD stores data from retired databases, namely the Pathway Interaction Database (Schaefer et al. 2009), and the Genetic Association Database from which the last data downloads were in 2012 and 2014, respectively (Becker et al. 2004).

The user interface

MOET is accessible from the RGD front page from “Analysis & Visualization” in the toolbar menu or as one of the tools embedded into RGD's individual Disease Portal pages from “Diseases” and “Disease Portals” (Fig. 2, A and B). MOET’s front page allows the use of an input list of genes or proteins in numerous identifier types with the desired species and ontology selection (Fig. 2C, Supplementary Fig. S1A1–A3). Alternatively, a genomic region of interest with the applicable assembly can be entered (Supplementary Fig. S1A4).

Open in a separate window

Fig. 2.

Access to MOET from A) and B) “Analysis & Visualization” and individual disease pages from “Diseases” and “Disease Portals”; C) You can enter your gene or protein list in the MOET interface as one of the acceptable RGD identifier types and then click on continue to view the results.

Visualization of output and downstream analyses

The results are displayed on the results page as a downloadable list (Fig. 3, A–C, Supplementary Fig. S1. B and C) and a graph. The functionality to perform comparisons across species and ontologies is provided by clicking on the species or ontology from the top of the results page (Supplementary Fig. 1C). Also, the P-value limit can be adjusted (Fig. 3D) with the available choices (0.01, 0.05, and 0.1) to modify only the displayed graph results accordingly (Supplementary Fig. S1C3A). The graph can be explored using the self-explanatory buttons at the top of the graph. Further, the user can send the list of genes to the other RGD analysis tools with the “All Analysis Tools” button and the resulting toolbox options (Supplementary Fig, S1B3).

Open in a separate window

Fig. 3.

MOET results page with downloadable list and graph; A) A list of over-represented terms with the uncorrected and Bonferroni-corrected P-values and odds ratio for each term; B) The species or ontology used for the analysis can be changed with the click of a single button. Current display is for rat and disease as indicated by the green tabs. Click Human and Pathway Ontology to obtain over-represented pathway terms for the list of human orthologs of the original input list of rat genes (refer to Supplementary Fig. S1); C) The result list can be downloaded. The resulting file contains the list of terms with the counts of genes annotated to each term in the input set and the reference set, the P-value, Bonferroni correction and odds ratio. D) The results shown in the graph of the number of annotated genes and P-value for each term in the result set can be made more or less stringent by changing the P-value limit using the drop-down list of options.

The number of annotated genes (Supplementary Fig. S1B2) beside the term in the table opens a pop-up containing the list of genes annotated to that term that can in turn be explored with MOET using the “Explore this Gene Set” link. The “All Analysis Tools” button and the toolbox can be used here to further analyze these genes with the other RGD tools.

Additional means to reach MOET are from the individual Disease Portals from “Diseases” in the toolbar at the top or “Disease Portals” in the RGD front page panel menu (Fig. 2, A and B). The user can select the required species and ontology category (Supplementary Fig. S2A) from the individual Disease Portal page (Kaldunski et al. 2021). Each Disease Portal has the associated gene, strain and QTL data integrated with it (Supplementary Fig. S2B). The “Gene Set Enrichment” section at the bottom of the disease page is integrated with MOET and sends the list of genes to MOET for analysis (Fig. 4, Supplementary Fig. S2C). Other ontologies can be selected from the bottom of the page below “Gene Set Enrichment” to analyze the list of genes annotated to the selected term for a different ontology.

Open in a separate window

Fig. 4.

MOET is accessible from individual Disease Portal pages. Here “Obesity/Metabolic Syndrome Portal” is shown; A) Rat as species and Disease as Ontology Category are selected as default; B) Number of genes associated with the selected species and disease category annotated to the term “familial hyperlipidemia” are shown; C) You can interchangeably select a different ontology for MOET analysis from the buttons below “Gene Set Enrichment.” Here, ontology analysis results for Rat in Chemicals and Drugs or ChEBI ontology are shown with “unsaturated fatty acid” as the top term.

Software comparison

To demonstrate the utility of MOET results a use-case was constructed from an experimentally derived gene set. Several popular enrichment tools were also selected to analyze these data and generate a results comparison. The software programs used in the comparison were MOET, DAVID (Huang da et al. 2009b), GSEA (Subramanian et al. 2005), and PANTHER (Mi et al. 2021). To generate a comparison that was consistent across the programs, several limiting factors were selected to maintain uniformity. An experimental gene set (Wang et al. 2015; GSE50027) of differentially expressed genes (DEG) from RNAseq analysis of liver samples from Lyon Hypertensive (LH/MavRrrcAek; RRID: RGD_10755352) and Lyon Normotensive (LN/MavRrrcAek; RRID: RGD_10755354) rats was used as a test gene set. These animals serve as a control (LN) and a disease (LH) model exhibiting many characteristic phenotypes associated with metabolic syndrome including hypertension, obesity, and dyslipidemia (Dupont et al. 1973; Vincent et al. 1993). The unfiltered DEG set began with 630 genes. The gene set that was submitted to each program was represented by Ensembl gene identifiers from the original published data (Wang et al. 2015). A common, minimal gene set of 411 genes recognized by all software was generated and used in the subsequent analyses (Supplementary Table S2). To maintain consistency, rat was selected as the species for genetic reference in all programs, and analyses were limited to intersecting curated canonical pathways and ontologies between the tools (Fig. 5). While each software included a pathway analysis, the resource data differed among them. As a result, limiting vocabularies to those common amongst all programs left only GO sets (Biological Process, Molecular Function, and Cellular Component) for inclusion in the analyses (Fig. 5).

Open in a separate window

Fig. 5.

Intersecting curated canonical pathways and ontologies used in software comparison. This Venn diagram depicts common and unique resources used by MOET, PANTHER, GSEA, and DAVID for integration into their respective ontology and pathway analyses. Several resources including KEGG and UniProt are included in the development of MOET ontologies, but results to their specific terms are not provided from the analysis.

Results

Software analysis comparison

The experimental common minimal gene set of 411 genes (Supplementary Table S2), differentially expressed between liver RNA from LH and LN rats, was used for this analysis to demonstrate a use case for MOET while comparing it with common ontology analysis tools, including PANTHER, GSEA and DAVID. The top 100 GO terms listed by ascending multiple testing corrected P-values were included in the results from each program. The resulting terms were compared and were included in Supplementary Table S4 if the term occurred in MOET and at least 1 other program’s top 100 list (not necessarily significantly enriched terms). The table displays the term name, GO ID, ontology name, software, and multiple testing corrected P-values. The list is ordered first by the number of programs containing the term, and then sorted by MOET P-values. Figure 6 gives an overview of the term overlap between software analyses. The MOET analysis results overlap with at least one other tool in 73 of its top 100 terms. Its greatest correspondence between tested software was with the PANTHER analysis tool (63 terms) followed by GSEA (27 terms) and DAVID (6 terms) (Fig. 6). There were 2 terms in common among all 4 programs (endoplasmic reticulum, GO:0005783; innate immune response, GO:0045087), 19 terms in common among 3 programs, and 52 terms in common between any of the 2 programs (Supplementary Fig. S3 and Table S4). Following the 2 terms in common between all programs, 5 of the next 10 terms have a high representation of terms associated with metabolic function (Supplementary Fig. S3 and Table S4). The cumulative nonoverlapping terms from the top 100 lists of each program represented 199 terms. Of these terms, 78 had multiple testing corrected P-values below 0.05 with 71 of them having corrected P-values below 0.01 (Supplementary Table S5). Terms unique to the MOET analysis accounted for 27 of the 199 terms and 7 of these terms had corrected P-values less than 0.05 (Fig. 6, Supplementary Table S5). An additional assessment on the top 20 GO Biological Process specific terms from MOET analysis had ten overlapping terms that occurred in the top 20 list of the other programs (Supplementary Table S6 and Fig. 7). Our results showed that MOET results are generally comparable and consistent with the commonly available ontology analysis tools. However, each tool (including MOET) also has differences that set them apart from one another. These differences could be attributed to some of the exclusive MOET features, and dissimilarities in statistical formulae, multiple testing corrections, ontologies, species, gene identifiers, and frequency of data updates between the tools. We have described the factors contributing to the differences in the Discussion section of this manuscript.

Open in a separate window

Fig. 6.

Representation of top 100 GO term overlap among compared enrichment tools. Gene Ontology analysis was performed using an experimentally derived DEG set. Genes were loaded into each software (MOET, PANTHER, GSEA, and DAVID). The top 100 terms ranked on P-value were assessed and overlaps are depicted in the Venn diagram. MOET has 63 terms in common with PANTHER, 27 terms with GSEA, and 6 terms with DAVID.

Open in a separate window

Fig. 7.

MOET top 20 GO biological process term overlap with compared enrichment tools. MOET contained 6 additional terms from the top 20 that were found in the top 100 terms from results in comparison software.

Discussion

GO over-representation analysis is an effective way to facilitate the analysis and interpretation of large amounts of -omics data. MOET, developed at RGD, is an ontology analysis tool that implements its assessment utilizing a web-based application. It provides 6 different ontology analyses with all the RGD species in an intuitive and user-friendly manner aiming for ease of use for researchers, particularly those without an extensive computer science background. MOET doesn’t need installation or preparation of local databases for its use. MOET, therefore, can utilize its integrated resources to facilitate novel, functional interpretation of user gene sets.

This introduction to MOET has highlighted many of the distinctive features that establish it as an easily accessible tool that provides unique analysis across multiple ontologies and species. Characterization of the tool with other popular tools commonly used for enrichment analysis has demonstrated consistency amongst results when using a benchmark gene set while providing a unique pattern of enriched terms. The comparison provided in this work used a gene set derived from DEGs found between a rat strain representing a model of metabolic syndrome (LH) and a control strain (LN). MOET generated overlapping results with established currently available tools and produced annotated term results from the GO which support a metabolic role for the differentially expressed gene set. It can be concluded that there is a good overlap with significant terms between MOET and other ontology analysis tools.

In our comparison, in addition to the commonality between terms, we also found differences in the number of annotations and P-values between the tools used for comparison. One source of difference could be that DAVID, PANTHER, GSEA, and MOET are based on different algorithms and use different methods for multiple testing corrections (Supplementary Table S1). Differences in the number of annotations and P-values between these tools can be attributed to some of the benefits that are unique to MOET. The continuing updates in ontology and annotations cause differences in significance values as new parent-child relationships increase the number of annotations to a term. Since MOET draws its underlying data directly from the RGD database, which is updated on a weekly basis, it has the most up-to-date ontologies and annotations resulting in the most accurate significance values. Another unique feature of MOET is its algorithm that includes only the genes in the selected species annotated to the selected ontology as the reference set. Thus, the P-value calculation is more precise compared to other tools which consider the entire gene list (global reference, e.g. DAVID) or require a user input a reference list.

The purpose of developing MOET was to provide the research community with a means to interpret experimentally derived gene sets. The breadth and depth of any scientific discipline and any individual researcher inherently have gaps in understanding and experience. The ontologies and species chosen for MOET are specifically designed to generate coverage for these gaps and produce functionally interpretable results. The description of MOET’s operability along with the use case provided in the results above establish a potential workflow to enable functional characterization of user-generated gene sets. An indication of support from the research community can be seen through MOET's usage since its first public release in April 2019. The stand-alone tool has been directly accessed over 9,000 times (September 2021 Google analytics query) and this count is likely an under-representation since MOET is also embedded into the Disease Portals which are not included in Google analytics counts.

RGD continues its commitment to providing the best in data and software tools for the research community. Future updates in MOET will include support for enrichment analysis that incorporates expression results and implementation of additional algorithms for P-value calculation. We also plan to integrate the option of showing a negative correlation between the genes and their respective annotated terms. We value feedback from the research community and strive to incorporate input and comments from users that assist in our software navigation and functionality. Each page in RGD has a link to send feedback or feedback can be submitted in the “Contact Us” form (https://rgd.mcw.edu/contact/index.shtml) at RGD.

Data availability

The authors state that all data necessary for confirming the conclusions presented in the article are represented fully within the article. MOET is freely available at https://rgd.mcw.edu/rgdweb/enrichment/start.html. MOET source code and documentation are available from Github at https://github.com/rat-genome-database/rgd-web-application/tree/master/web-app/WEB-INF/jsp/enrichment.

Supplemental material is available at GENETICS online.

Supplementary Material

Acknowledgments

Literature cited