COFECO: composite function annotation enriched by protein complex data.

COFECO is a web-based tool for a composite annotation of protein complexes, KEGG pathways and Gene Ontology (GO) terms within a class of genes and their orthologs under study. Widely used functional enrichment tools using GO and KEGG pathways create large list of annotations that make it difficult to derive consolidated information and often include over-generalized terms. The interrelationship of annotation terms can be more clearly delineated by integrating the information of physically interacting proteins with biological pathways and GO terms. COFECO has the following advanced characteristics: (i) The composite annotation sets of correlated functions and cellular processes for a given gene set can be identified in a more comprehensive and specified way by the employment of protein complex data together with GO and KEGG pathways as annotation resources. (ii) Orthology based integrative annotations among different species complement the defective annotations in an individual genome and provide the information of evolutionary conserved correlations. (iii) A term filtering feature enables users to collect the specified annotations enriched with selected function terms. (iv) A cross-comparison of annotation results between two different datasets is possible. In addition, COFECO provides a web-based GO hierarchical viewer and KEGG pathway viewer where the enrichment results can be summarized and further explored. COFECO is freely accessible at http://piech.kaist.ac.kr/cofeco.

INTRODUCTION

High-throughput experiments such as microarrays, serial analysis of gene expression (SAGE), chromatin immunoprecipitation (ChIP)-on-CHIP and proteomics generate a number of interesting gene sets that are functionally correlated within a certain biological condition. For an interpretation of the functions and biological processes for gene sets under study, enrichment based functional annotation is a common and suitable method. Various enrichment tools have been developed and are widely used but some challenging issues still remain unresolved (1–12). Annotation databases need to be extended for the comprehensive identification of biological processes for a gene set of interest. In addition, the interrelationship of heterogeneous annotations should be integrated in order to make functional annotations more interpretable within a network context. Currently, enrichment tools contain various biological annotation resources such as Gene Ontology (GO), Pfam domains, InterPro motifs, KEGG pathways and so on. However, protein complex information has not been used extensively for enrichment resources. The physical interactions of co-complexed proteins support a solid basis for assigning correlated proteins working together for specific functions. Other various functional annotations can be associated in a protein complex as correlated functions. Protein complex data show many similar variants that may reflect the dynamic changes of functional modules under various cellular conditions. Therefore, the interrelationship of cellular functions can be comprehensively and specifically delineated by integrating the information of a protein complex with other functional annotation resources. To obtain the integrated functional annotations from heterogeneous resources, composite annotation methods can be applied (8,9,10). All annotation terms including concurrent genes are composited and evaluated in order to provide the best composite annotations for the given gene set. The interrelating feature of a composite annotation algorithm can be combined synergistically with protein complex annotations. For the same purpose, a protein interaction network, alone or together with complexes, can be suggested but the functional boundary of an interaction network is ambiguous and still suffers from a high false-positive rate mainly due to the wrong interpretation of co-complex data (13). COFECO is a web-based tool that improves on the aforementioned issues in current enrichment tools by using a composite function annotation with protein complex data, KEGG pathways (14) and GO terms (15) for a given set of genes and their orthologs. COFECO enables comparative analyses between different gene sets with cross comparison tool. The correlated functions for an annotated complex can be conveniently explored at the GO and KEGG pathways using graphical viewers. We combined wide-spread protein complex datasets (15–22) so that it covers a large enough number of proteins and annotations so as to be comparable to other annotation resources (Table 1).

Table 1.

Statistics of annotations and proteins in annotation resources of COFECO

Organisms	Proteins					Annotations
	Complex	KEGG	GO BP	GO MF	GO CC	Complexa	KEGG	GO BP	GO MF	GO CC
M. musculus	5670	5313	33 714	39 440	34 231	2118	199	3584	2320	614
S. cerevisiae	5246	1237	5147	5020	5761	8049	110	1556	1484	511
H. sapiens	4123	5507	36 780	40 115	36 594	2858	205	3082	2612	684
R. norvegicus	2957	2674	9407	10 729	9837	1850	197	2059	1858	463
G. gallus	1820	604	4379	5058	4345	1287	121	894	984	289
A. thaliana	1606	1947	19 694	25 962	13 380	619	132	1043	1240	269
D. melanogaster	1583	2216	13 862	17 892	10 666	865	135	2196	1545	439
S. pombe	899	966	3553	3429	4790	526	106	1153	1082	370
C. elegans	806	1012	9891	11651	6456	600	122	1163	969	237

aThe numbers of annotations for complexes are the number of complexes for each genome. Among the whole 19 064 complexes, 63% have their own function annotations or complex names and 37%, which are from high-throughput co-complex data in general, do not have complex name. In this system, complex itself is used as an annotation unit representing functionally correlated group.

MATERIALS AND METHODS

Inputs

A list of gene sets can be the input into COFECO. Gene (or protein) identifiers from various databases including UniProtKB, iProClass, Entrez Gene, UniGene, RefSeq, EMBL, ENSEMBL, SGD, RGD, MGI, HGNC and IPI are allowed. COFECO also accepts microarray probe identifiers of Affymetrix and Agilent.

Data resources

The protein complex datasets employed in COFECO are as follows: complex terms specified within the GO cellular component category (15), CORUM (16), Reactome (17), MPact (18), PINdb (19) and three high-throughput TAP/Mass datasets (20–22). The statistics of protein complexes completely included in three GO categories and KEGG pathways is as follows: 7737 (40%) in GO biological processes, 8767 (46%) in GO cellular components, 8693 (46%) in GO molecular functions, and 7860 (41%) in KEGG pathways, respectively. Those complexes create new complex-subclasses in GO or KEGG pathways, which more specify or cross-correlate the classes in both resources. 6431 (34%) complexes have proteins both included and not-included in GO or KEGG. Especially, 1473 (8%) complexes has completely new members that are not included in GO or KEGG. The complex datasets support the following 21 organisms: Homo sapiens, Mus musculus, Rattus norvegicus, Arabidopsis thaliana, Drosophila melanogaster, Saccharomyces cerevisiae, Schizosaccharomyces pombe, Caenorhabditis elegans, Dictyostelium discoideum, Gallus gallus, Bos taurus, Mycobacterium tuberculosis, Danio rerio, Xenopus laevis, Sulfolobus solfataricus, Sus scrofa, Canis lupus familiaris, Xenopus tropicalis, Methanocaldococcus jannasc, Synechocystis sp. PCC 6803 and Pan troglodytes. KEGG pathways, GO terms and other annotation files were downloaded from public ftp servers. The orthologs of query genes in user-specified organism are acquired from the eukaryotic ortholog database, InParanoid (23) that is generated by using orthologs and in-paralogs detection algorithm. Ortholog set is analyzed separately from original query gene set. More details on the data collection can be found in supplementary material.

Composite enrichment algorithm

COFECO uses a modified form of the a priori algorithm (24) for composite enrichment that generates sets of associated annotations which co-occur significantly in a set of genes. Composite enrichment algorithm consists of two processes: generation of composite annotation terms and statistical evaluation of them. In the generation of composite annotations, the a priori algorithm begins by selecting the set of all single annotation terms that occur in at least k concurrent genes. In the next step, two terms that occur in at least k concurrent genes are merged to a new associated term. The process continues until the longest associated terms are found. In the statistical evaluation process, composite annotations which are significantly enriched in a given gene set are evaluated. As the number of annotation resources increases and k decreases, the computational complexity might drastically grow to enumerate all possible compositions of annotation terms. In addition, the protein redundancy among complexes may also lead to huge computation in COFECO analysis. To solve this problem, we developed the greedy algorithm that select the top ranked K terms determined by P-value calculations at each step of composite annotation. The greedy algorithm is optionally applicable to the composite annotation depending on user's preference. More details on the greedy algorithm can be found in supplementary material. A statistical significance test is applied to all single and associated terms found in the above process. A hypergeometric distribution, binomial test, Fisher's exact test, or chi-squared test can be applied in COFECO. The multiple testing correction of P-value can be conducted using Bonferroni correction, the Holm–Bonferroni method, or a false discovery rate (FDR) method (25). In COFECO, these processes are implemented as both single and composite enrichment of various combination of annotation resources can be performed simultaneously or selectively. COFECO implements special composite annotation processes. First, annotation resource combination can be performed by four types: mandatorily including protein complex and at least one different resource, mandatorily including protein complex, mandatorily including at least two different resources, and all possible combinations. These operations support term-term associations by considering annotation resources and specify biological annotations. In addition, COFECO provides term filtering function that enables the enrichment of selected annotation terms with user-specified keywords.

Outputs

COFECO reports a summary of annotation result, composite annotation, single annotation and details of enriched protein complex for requested genes and their orthologs. A summary table provides the frequency of enriched annotation terms and associated gene sets. A composite annotation table provides a list of annotations, their associated genes, P-value and public website links for the annotation resources. A single annotation table displays typical enrichment results of individual resources without term association process. An enriched protein complex table shows all available information of the complex including KEGG pathways and GO terms. KEGG pathways and GO terms in an enriched annotation table and a protein complex table are summarized by web-based viewers. COFECO outputs are accessible at specified URL addresses that are notified by an Email and can be used as an input file of cross-comparison analysis in COFECO.

Implementation

COFECO was implemented on Linux and runs on Apache Web Server combined by a Tomcat servlet engine. A composite annotation algorithm was implemented in Java to take advantage of serialization, reusability of data objects and platform independence. Java serialization supports much faster and simpler manipulation of output objects in different processes such as cross comparisons or output reporting. All preprocessed data used within our system were stored in Oracle 9i DBMS. COFECO web pages were developed with Java Server Page and tested with most available web browsers. A GO hierarchical viewer was implemented in a Java Applet and JUNG library at http://jung.sourceforge.net. KEGG pathway viewer was developed using the open web services of KEGG at http://soap.genome.jp/KEGG.wsdl. More organisms, annotation resources and identifiers will be systematically updated regularly.

Functionalities and characteristics

The details of functionalities and characteristics of COFECO are as follows.

Employment of protein complexes as an annotation resource

The addition of protein complex within annotation resources means more than simple expansion of annotation space. Protein complexes show the precise composition of collaborative proteins for specific functions under various cellular conditions in different time and locations. The conservation and variation of protein members and corresponding functions in complex give the information of dynamic cross correlation among cellular functions and processes. In these senses, the composite annotation of protein complexes with other annotation resources such as GO terms and KEGG pathways provides more specified protein groups with comprehensively correlated functional contexts of GO and KEGG within complex unit.

Comparative analysis with orthologs

A unique feature of COFECO is orthologs-based annotation analysis. COFECO performs composite or single annotations of ortholog set in a user-specified organism and reports the result separately from original query genes. User can acquire putative complementary annotations from different organisms with the insight of evolutionary conserved or differentiated functional groups in a given gene set by comparing the annotations for queried genes and orthologs.

Summarization and exploration of annotated functions via intuitive graphical views

The hierarchical relationship of enriched GO terms can be summarized efficiently by a web-based GO viewer, that GO terms are color-marked by the enrichment types or your selection (Figure 1C and D). KEGG viewer has color-marking function for a significant set of genes which are co-annotated with protein complex and KEGG pathways. The enriched members with the other co-complexed proteins are marked in the KEGG pathway (Figure 1E and F).

Figure 1.

Screenshot depicting the results of the analysis of 85 differentially expressed genes (DEGs) in human testis tissue. (A) Composite annotation results from the function annotation of protein complexes, KEGG pathways and GO biological processes. (B) Composite annotation results of an ortholog list of 85 DEGs. (C) and (D) Graphical GO views of A and B, respectively. (E) Inset of KEGG cell-cycle pathway, which involves co-complexed proteins of cell cycle kinase complex. (F) Inset of KEGG cell-cycle pathway, which involves co-complexed proteins of kinetochore. Red outlined rectangles in Figure 1E indicate co-complexed proteins of the protein complex named as ‘centrosome containing phosphorylated Nlp’, which is actually ranked the third annotation, not displayed in Figure 1B. Yellow boxes indicate statistically significant genes in enriched composite annotations and pink boxes indicate co-complexed proteins except for statistically significant genes. Dotted lines indicate same genes.

Specified composite annotation via term filtering

User can optionally specify the annotation terms to be included or excluded in the enrichment by setting the keywords for the annotation terms.

Cross comparison between the annotation outputs

A cross-comparison is used to identify changes/trends between the annotation results of two different datasets. This functionality is useful to compare various types of outputs, for examples, enriched annotations of input list of genes with those of ortholog list of input genes, and annotation outputs from different input sets.

An example of COFECO analysis

Figure 1 summarizes the functionality of COFECO with an example of 85 human testis-specific genes (26) that has been used previously for the test of other enrichment tools (3,8). The previous composite annotation analysis (8) could successfully point out new explicit connection between the terms of ‘protein amino-acid phophorylation’ and ‘cell cycle’ out of large categories annotated by single annotation analysis (3). The new connection was interpreted with five best composite annotations from GO biological process and InterPro motifs. In a single and composite annotation with typical analysis setting with GO and KEGG pathway, COFECO showed basically same results as previous studies. However, composite annotation based on complex and orthology summarized all previous conclusions and more specific annotations from the first ranked composite annotations (Figure 1A and B). From a few composite annotations with the significant P-value, more specified annotations and gene sets could be summarized on complex units with comprehensively correlated functional contexts of GO and KEGG. The relevance of enriched annotations could be confirmed by the known protein complex information. For example, cell cycle related kinase complexes of the first annotation shown Figure 1A perform the control of the cell cycle at the G2/M (mitosis) transition through cyclin-dependent kinase activity (27). Kinetochore related complexes of the second annotation have kinase activity and interact with centromere and spindle during cell division (28). The importance of orthology-based composite annotation is shown in Figure 1B. The GO and KEGG annotations in human and mouse were complementary and informative, especially with specific terms like ‘p53-signaling pathway’ and ‘protein amino-acid phosphorylation’. The annotation results of mouse also suggest an additive gene of interest such as, ARK1, which is serine/threonine protein kinase 6 and is involved in microtubule formation/stabilization. There were interesting observations related to human protein complex, ‘cell-cycle kinase complex CDC2 complex’, which consists of six proteins: CCNB1 (CycB), CCNB2 (CycB), CDC2, CCND1 (CycD), CDKN1A (Cip1) and PCNA. KEGG cell cycle pathway completely contains these six co-complexed proteins (Figure 1E). CCND1, CDKN1A and PCNA which were not involved in the enriched genes could be inferred as importantly correlated genes by this integrated analysis. Figure 1E and F provide insight into the associated relationships among three annotated protein complexes: ‘cell cycle kinase complex’, ‘kinetochore’ and ‘centrosome containing phosphorylated N1p’ (third annotation in mouse which is not shown at Figure1B) with KEGG viewer summary. Fourteen proteins [CycB, CycD, CDK1, CDK2, Cip1, PCNA, Pik1, 14-3-3, Bub1, Bub3, BubR1 (BUB1B), Mad1 (MAD1L1), Mad2 (MAD2L1) and Cdc20] of the three protein complexes above were annotated in KEGG cell cycle pathway. These proteins can be highlighted as significant gene sets that narrows down the scope of KEGG cell cycle correlated with a human testis-specific expression. Another example showing the unique functionalities and features of COFECO can be found in supplementary material.

CONCLUSION

A large and linear list of enriched annotation terms in the outputs of functional enrichment tools is often incomprehensive or irrelevant toward understanding the biological functions of a gene set under study. Here, we present COFECO, a web-based tool for the composite annotation of protein complexes, KEGG pathways and GO terms within a class of genes and their orthologs under study. As has been illustrated in an example, the composite annotation of protein complexes with other annotation resources such as GO and KEGG pathway provides more specified annotations and gene groups with comprehensively correlated functional contexts of GO and KEGG within complex unit. Our tool can also furnish additional proteins of interest among co-complexed proteins. In addition, comparative analysis with orthology and a cross comparison between annotation outputs provide interesting phenomena that are not addressed by the annotation outputs of a single dataset. The aforementioned features make COFECO a useful tool for users discovering the biological functions of experimental data.

SUPPLEMENTARY DATA

Supplementary Data are available at NAR Online.

FUNDING

Funding for open access charge: Development of Next-Generation New Technology program (10024715-2008-21).

Conflict of interest statement. None declared.