Fourmidable analyses are carried out via custom Perl scripts and publicly available software. Annotation and assembly information is stored in a MySQL database while sequences and BLAST 16 results are kept in indexed text files for rapid retrieval while limiting database size. Data are stored separately for different species [see Additional file 1]. Computationally intensive tasks are parallelized on the Vital-IT high-performance computing cluster [17, Additional file 2]. Fourmidable should thus readily handle large amounts of additional data. Data is accessible to web users via a PHP/Apache-based interface hosted by the Swiss Institute of Bioinformatics.

Fourmidable currently processes nucleotide sequences for the red fire ant Solenopsis invicta and the black garden ant Lasius niger. When available, raw .ab1 or .scf trace files are converted via Phred 18 to FASTA nucleotide and quality score files. Additional sequences for which trace files cannot be obtained are downloaded from Genbank; quality score files are generated for these sequences with an arbitrary Phred quality score of 25. All input sequences are cleaned using Lucy 19, DUST [Tatusov and Lipman, unpublished], RepeatMasker 20 and CrossMatch 21 to respectively remove low-quality regions and sequences, low-complexity regions, interspersed repeats, and sequences from bacteria, organelles or cloning vectors. Cleaned sequences are then compared via reciprocal BLAST 16, and subsequently similar sequences are grouped into clusters. Within each cluster, sequences are independently assembled via CAP3 22. This circumvents memory constraints that CAP3 would face if attempting a global assembly with large numbers of sequences. The output from clustering and CAP3 assemblies are contigs (each is the consensus of several assembled sequences) and singlets (sequences that did not assemble with others). All sequences are subsequently annotated as follows.

All sequences are compared to the non-redundant protein database 23 via BLASTX as well as to several insect-specific databases via TBLASTX, BLASTX and BLASTN. Gene Ontology (GO) annotation of the strongest of the top five BLASTX hits to the non-redundant protein database is carried over to ant sequences.

To determine possible peptide sequences, we compute all six possible translations of transcriptome sequences with the potential to encode sequences longer than 30 amino acids. We do not use an ab initio gene prediction program such as ESTScan 24 because the sensitivity of such programs is limited by the absence of solid training data for ants. Instead, all potential open reading frames are annotated via Interproscan 25. Some Interproscan hits directly provide GO annotation of ant sequences, complementing the BLASTX-inferred GO annotation mentioned above.

The BLAST, Interproscan and GO annotations are updated every two months or when new sequence data is added to the assembly pipeline.

At several steps during this assembly and annotation pipeline, bioinformatics software was run with parameters that differed from default parameters [see Additional file 3], as determined by using reduced test datasets.

Fourmidable is linked to the GEDAI gene expression database [Robin Liechti, unpublished]. Storage and simple analysis using Bioconductor packages 26 is possible for data from single-color and two-color spotted microarrays, as well as for Affymetrix and Illumina microarrays.

There are several manners of accessing sequence information in Fourmidable. First, single sequences can be searched by species as well as by partial identifiers for input sequences or assembled contigs. Second, lists of identifiers can be used for searching. Third, user-supplied sequence data can be used for BLAST similarity searches against sequences in Fourmidable. Finally, users can navigate inferred Gene Ontology annotations for biological processes, cellular components, and molecular functions using the AmiGO browser 28. The first two search manners result in tables as described below. BLAST searching and GO browsing produce lists of sequence identifiers that can be used as inputs for the first two search manners.

Sequence searches result in tables with one line per sequence for easy access to sequence annotation (see Figure 1). In particular:

- Clicking on an identifier in the "Raw Sequence" column provides information on how that sequence was obtained and allows users to download the raw sequence. Tracefiles can be viewed with the Baylor College of Medicine Trace Viewer 29 or downloaded.

- Cleaned FASTA-format sequence can be downloaded for individual singlets and contigs.

- If the sequence is part of an assembled contig, the "Assembled Contig" identifier links to a display of the consensus sequence and the relevant input sequences as well as their quality scores. Additionally, a multiple sequence alignment highlights nucleotide polymorphisms within the consensus sequence.

- BLAST results between the sequence of interest and sequences from the non-redundant protein database and several insect nucleotide and protein databases are summarized by blue and red buttons, indicating weak (0.01 > E-value > 10^-5) and stronger similarity (E-value ≤ 10^-5) respectively. Clicking on a button displays the complete BLAST report.

- An additional link to Interproscan results and six-frame protein translations is displayed if Interproscan annotation is available.

A convenient repository is available for ant molecular biology protocols (commonly in .doc or .pdf formats). New or revised protocols can be added via an upload form. Fourmidable also supports upload of result files from microarray gene expression experiments. The GEDAI platform allows straightforward sharing of microarray results and performing simple microarray analyses (including preprocessing, direct and indirect two-sample comparisons, 2 × 2 factorial and gene set enrichment analyses). GEDAI also provides summaries of the expression levels of specific microarray probe identifiers across multiple microarray experiments. Finally, Fourmidable provides download links to individual files containing all raw or assembled sequences, as well as sequence annotation in text format [see also Additional file 4].

The sequence assembly and annotation information provided by Fourmidable has already proved useful in several published studies 1430. Most recently, Fourmidable's data helped J. Wang and colleagues to characterize genes that are differently expressed between workers from two alternative social forms of fire ants 6.

Fourmidable was initially developed as a private database in Lausanne. Recently it has been updated and made publicly accessible because of increasing interest in ant molecular research. To further develop Fourmidable, several primary investigators in the USA have submitted grant applications. This should lead to improved integration of gene expression data with sequence annotation, as well as support for genetic mapping and linkage data. When large amounts of genomic sequence become available for ants, the current approach for assembly and annotation may become computationally unrealistic. An alternative may be to adapt existing genome assembly, annotation and browsing tools.