Four different types of primary data are obtained from a variety of repositories and integrated into the e-Fungi database: genomic data, Gene Ontology annotations, pathway data and EST data. Genomic data consists of the genome sequence with varying degrees of annotation. This annotation can include the prediction of genes with their introns, exons and predicted proteins, as well as their locations on contigs, supercontigs or chromosomes. Table 1 lists all the genomes integrated into the e-Fungi database with the data sources from which the data has been obtained. Other data has been obtained as follows:

• Gene Ontology annotation for S. cerevisiae, S. pombe and C. albicans has been obtained from SGD 11, Sanger GeneDB 26 and CGD 27.

• Pathway information including the assignment of pathways to proteins for S. cerevisiae, S. paradoxus, S. mikatae, S. bayanus, E. gossypii, K. lactis, K. waltii, D. hansenii, C. albicans, C. glabrata, Y. lipolytica, S. pombe, N. crassa, M. grisea, A. nidulans, A. fumigatus, A. oryzae and E. cuniculi has been obtained from KEGG 17.

• Expressed Sequence Tag (EST) data are obtained from the COGEME Phytopathogenic Fungi and Oomycete EST Database 28.

The following kinds of derived data are stored in the database:

• Clustering sequences from 36 fungal genomes: We compared 348,995 protein sequences from the 36 genomes integrated in e-Fungi (see Table 1) using BlastP 29 with an E-value cut-off of 10^-5. This resulted in 47,342,483 hits. Markov Chain Clustering (MCL) 30 was then applied to generate clusters of similar proteins, using 2.5 as a moderate inflation value and 10^-10 as a comparatively strict E-value cut-off. This generated 23,724 clusters containing in total 282,061 sequences, while 66,934 sequences were singletons.

• Orthology assignments: To identify orthologous proteins between the 36 genomes, the BlastP results were analysed with OrthoMCL 31 using its default parameters (i.e., an E-value of 10^-5). The analysis produced in total 30,084 clusters, with 5,406 of those containing just paralogues and 24,678 containing potential orthologous proteins. Out of these clusters of potential orthologues, 14,113 are unambiguous orthologue clusters, while 10,565 are ambiguous clusters with orthologues and recent paralogues.

• Domain assignments: To identify functional domains and other known sequence motifs, predicted proteins from all 36 genomes were scanned with the Pfam database release 18 16 using hmmpfam 32. A total of 6,260 different Pfam domains were identified in 196,425 proteins, using an E-value cut-off of 0.1. The distribution of 5 of the most frequently found Pfam domains among the genomes is shown in Figure 2.

• Protein localisation predictions: Protein sub-cellular localisations were predicted using SignalP 33, PSort 34 and Wolf-PSort 35 with the default parameters. Distributions of the most frequently assigned PSort and Wolf-PSort predictions among the genomes are shown in Figures 3 and 4.

All the generated data are integrated into the e-Fungi database using the loading infrastructure described below.

The Object Database Management System Versant FastObjects 36 is used to store the data integrated into e-Fungi. The database schema has been implemented using Java Data Objects (JDO), an industry standard interface-based abstraction of persistence. Using JDO for storing the data allows the direct implementation of the object data model without the need to map between the object model and, for example, a relational database model. Such a mapping often results in a less intuitive representation of the data.

Using an object data model in combination with an object-oriented programming language, such as Java, also enables a tighter integration of analysis tasks with the stored data. Complex queries that analyse a large variety of different types of data can, therefore, be realised in a fairly intuitive manner.

The database schema can be divided into different parts, modelling the different types of data introduced above. The parts of the schema for genomic sequences, annotations, pathways and ESTs are based on published models 17283738. The part of the schema modelling the derived data is introduced in more detail in the following.

Results of the MCL and OrthoMCL cluster analyses consist of an identifier for each cluster and the assignments of proteins to clusters, captured in the classes MclCluster and OrthoMclCluster. To be able to retrieve the MCL cluster or the OrthoMCL cluster for a particular protein, the class Protein has an association with both MclCluster and OrthoMclCluster.

The results of the predictions of protein sub-cellular localisations are captured following a similar approach for all three different prediction methods. Each prediction method can have a number of different outcomes, e.g., golgi, cytoplasmic, or plasma membrane. These are captured in PSortPrediction, WolfPSortPrediction and SignalPPrediction. Each prediction has a 0-to-many association with Protein, enabling the retrieval of all proteins with a particular predicted localisation. However, not only are the final predictions provided as a result of the analyses, so are a number of scores associated with the predictions. Scores returned by each prediction analysis are captured in PSortResult, WolfPSortResult and SignalPResult, which have a 1-to-1 association with the protein for which the prediction has been made. The scores are captured as provenance information associated with each analysis, thereby recording all the information contained in the report provided as a result of each analysis.

A loading infrastructure has been developed to integrate data from a variety of data sources, as listed in Table 1, and map the information onto the e-Fungi database schema. The infrastructure consists of 3 general modules, the loaders, parsers and wrappers (see Figure 6). The loaders are specific to each data source and data format. For example, the genomic data loader for data from the Broad Institute gathers the contigs, genes and protein sequence data from 3 separate FASTA files and relates the data to information provided in other files of different formats. Data from each source is parsed into a generalised format that can be processed by the loader using the respective parsers for each of the available data formats (e.g., FASTA, GTF, GFF). The wrappers are responsible for creating and linking objects, e.g., when loading data on a protein the wrapper will create a Protein, create a PrimaryPolypeptide sequence, and link these together.

The loading infrastructure is designed to minimise maintenance and ensure extensibility of the database. Each module protects the others from sections that are prone to changes. For example, changes in the source, e.g., changes in location, format or method of access, will only require modifications of the loaders without affecting the existing parsers or wrappers. The parsers, on the other hand, act as tools for the loaders and can be easily improved or added as required. For the database, wrappers provide a layer of protection that allows the schema to be changed without the need to modify existing loaders or parsers. This allows the database to be extended easily to include further kinds of data without the need to rebuild the loading infrastructure.

The data stored in the e-Fungi database can be analysed using pre-determined analysis tasks that can be parameterised, so-called canned queries. More than 90 queries are currently available, varying in their complexity from simple retrieval tasks to complex analysis tasks. Similar queries based on the type of data analysed are grouped together (see Table 2 for an overview of the categories). Providing pre-determined analysis tasks as a means to explore and analyse the stored data might seem quite limiting at first. However, as shown in the next section, it allows complex analysis tasks to be provided that are beyond simple keyword- or identifier-based retrieval of stored data.

Browse provides an overview of the genomes stored in the database, but can also be used as an entry point to explore and analyse the data by following either of the two different kinds of links provided: (i) navigational links, and (ii) links to analysis tasks. The former, for example, enable the retrieval of contigs or chromosomes for a particular genome, whereas the latter link to analysis tasks provided in the canned query library using the chosen entry, e.g., a particular genome, as input.

The Simple Search feature exposes the tasks provided in the canned query library mentioned above. Documentation for each query can be found under the category 'Documentation' in the menu on the left hand side of the e-Fungi web site or by following the 'Help' link provided on each query form (see Figure 7). Information includes the type of input required, a number of example inputs, and a description of the output provided by the query. Furthermore, information on the runtime of long running queries is also provided. The canned queries provided to support the comparative analyses listed in the background section are introduced in the following to illustrate the utility of the e-Fungi database:

1. Identification of species-specific protein families or those conserved in closely related species. The queries 'Get MCL clusters with proteins of a given genome' or 'Get OrthoMCL clusters with proteins of a given genome', which can be found in the group 'MCL cluster analysis' and 'OrthoMCL cluster analysis', respectively, can be used to retrieve all clusters containing proteins of a particular genome and perhaps identify clusters containing only paralogues of the chosen genome, i.e., possible species-specific proteins. To run a query, the appropriate query category is chosen, e.g., MCL clusters or OrthoMCL clusters. From the list of canned queries in the chosen group, the canned query of interest is selected and the user is presented with a form for the required input parameters (e.g., Figure 7). For some of the input parameters, an existing value feature exists, enabling users to choose a value from a list of possible values. For other parameters, the Advanced Search feature can be used to retrieve the exact value, such as for a particular Gene Ontology Annotation or Pfam domain, as illustrated later. With the input parameters provided, the query can be executed and the results displayed (see Figure 8). Some of the result reports provide navigation in the form of links, similar to the navigation in Browse, for further exploration and analysis of the results.

2. Contraction or expansion of protein families. The query 'Get all MCL clusters with more than a given percentage of proteins of a given genome' can be used to identify outlying clusters. The query is part of the group 'MCL cluster analysis' and has a counterpart in the group 'OrthoMCL cluster analysis'. To identify protein families that are conserved in genomes exhibiting a certain phenotype, the query 'Get MCL clusters containing proteins of a group of genomes' or its counterpart that analyses OrthoMCL clusters can be used. A group of genomes can be specified by their exhibited phenotypes, such as growth form or pathogenicity. Analyses to identify species-specific protein families or those that are conserved in related species with a particular phenotype, as well as studies of contraction or expansion of protein families, have been part of recent comparative studies 3404142.

3. Genome redundancy in a range of related species, illustrating the importance of genome duplication 4344. The canned query 'Get the number of paralogues for all clusters containing proteins of a given genome' that can be found in the group 'Redundancy analysis' can be used.

4. Identification of secreted proteins, which in pathogens could play important roles in host-pathogen interactions. This analysis can be aided by executing either of the following canned queries 'Get secretory proteins for a given genome', which is part of the group 'Secretome analysis', or 'Get annotation for proteins of a given genome in MCL/OrthoMCL clusters with secretory proteins'. The queries retrieving the annotation of proteins in MCL or OrthoMCL clusters are part of the group 'Annotation of proteins in clusters'.

5. Conservation of genes defined as essential for growth in Saccharomyces cerevisiae 8 among fungal genomes. This analysis is supported by a number of queries that can be found in the groups 'Essential yeast genes cluster analysis' and 'Essential yeast genes orthology analysis'. Similar studies using the essential genes identified in Candida albicans have been reported in 7.

6. Conservation of components of metabolic pathways among fungal genomes 45. Again, this analysis is supported by a number of canned queries, such as 'Get proteins that are in the same (KEGG reference) pathway as a given protein' of the group 'Pathway analysis', which retrieves all the proteins that are known to participate in a particular pathway. To analyse newly sequenced genomes and identify proteins that could potentially be part of a pathway, the query 'Get annotation for proteins of a given genome in the same MCL/OrthoMCL clusters as proteins in a given pathway', part of the 'Annotation of proteins in clusters' group, can be used.

The Advanced Search feature can be used to retrieve entries for which a property value or a range of property values can be specified. The user specifies the type of entry to be retrieved and the filters that the returned entries have to match. Similar to the Simple Search, a form is provided requesting input parameters for the Advanced Search. The example of an Advanced Search shown in Figure 9 retrieves all the biosynthesis pathways, i.e., all the KEGG pathways the name of which ends in 'biosynthesis'.