AnimalTFDB: a comprehensive animal transcription factor database

Data sources

Currently, AnimalTFDB contains TFs, transcription cofactors and chromatin remodeling factors identified in 50 animals (Table1). All genome data were downloaded from Ensembl (release version 60, http://www.ensembl.org/) database.

Animal TF family list and their HMM profiles

We characterized and classified TFs by their sequence specific DBDs. After reviewing literatures, we finally collected and curated 71 animal TF families and a group named ‘others’ including some orphan TFs (http://www.bioguo.org/AnimalTFDB/help.php), which is currently the most complete TF family list for animals. Among them, 59 families had Hidden Markov Model (HMM) profiles for their DBDs in Pfam database (v25.0) (13), while no HMM profiles were available for the other 12 TF families. We built HMM profiles for them based on their DBD multiple sequence alignments by the hmmbuild program in the HMMER package.

TFs identification

We applied the hmmsearch program in HMMER package to search all the protein sequences against the DBD HMM profiles to predict TFs. Based on our manual checking for the predicted human and mouse TF results, we took E-value 0.0001 as the cutoff, which simultaneously considered the accuracy and sensitivity. For TFs that had more than one DBD, we assigned them into families based on their true DBD, which is the domain exactly binding to DNA in those proteins.

Identification of transcription cofactors and chromatin remodeling factors

In AnimalTFDB, transcription cofactors were considered as proteins that interact with TFs in the transcription apparatus but are not able to bind the DNA directly. The chromatin remodeling factors were defined as proteins that regulate transcription by modifying the chromatin formation. To identify them, we firstly got the human transcription cofactors and chromatin remodeling factors from TFCONES (9) and Gene Ontology (GO) (14) databases according to the GO items: transcription cofactor activity and chromatin remodeling, respectively. Then, we used the human sequences to perform BLAST search and chose the best BLAST hits as the transcription cofactors or chromatin remodeling factors for the searched species.

Annotations of the identified factors

The numbers of TFs, transcription cofactors and chromatin remodeling factors identified in 50 animals were showed in Table 1. In order to provide more useful information, we made extensive annotations for them. We obtained the basic gene information and GO annotation from NCBI and Ensembl databases. Putative functional domains and 3D structure hits for the longest protein of each gene were offered. The protein–protein interaction information was parsed from BioGRID (15), HPRD (16) and An atlas of human and mouse TF interactions (17) databases. The pathway annotations from BioCarta (http://www.biocarta.com/) and KEGG (18) databases were available in AnimalTFDB. TFs binding sites and target genes were extracted from TRED (19) and JASPAR (20) databases. In addition, we also provided links to GenBank, Unigene and many species-specific databases such as: MGI, HGNC, FlyBase and so on.

Putative ortholog and paralog annotation

To predict the putative orthologs of these factors among different species, the reciprocal best hit (RBH) method (21) was used. We performed the all-against-all BLASTP search between proteins of two genomes with strict cutoffs E-value ≤ 1e–20, coverage ≥ 70%, identity ≥ 50% and set the reciprocal best hit pairs as orthologs. While, we applied the BLAST score ratio (BSR) (22) approach to predict paralogs. BLASTP search was done in each genome with the same benchmark applied in ortholog finding. After comparing the results of different BSR value, we chose the BSR value 0.4 as the cutoff for paralogs.

Database organization

Considering MySQL is a free database management system widely applied in bioinformatics, we stored all the information of AnimalTFDB in a MySQL database. Since the different TF annotations varied in contents and formats, we classified all the data into 30 separated tables. The Ensembl ID and Gene ID were used as the main keys to organize and link all the tables.

Data browse

To help users browse the data conveniently and clearly, AnimalTFDB provided two different ways to browse the data: (i) browse by species; (ii) browse by family. On the browse family page, all TF families were further merged into six groups based on the TRANSFAC classification: helix–turn–helix, other α-helix, zinc-coordinating, basic domains, β-scaffold and unclassified structure. The TF family list in each group was shown by the treeview on the left part of this page and the 3D structure images of TF DBDs were used as the family logos on the right part. On the browse species page, 50 species were classified into 11 categories according to the Ensembl taxonomy, which were primates, rodents, laurasiatheria, afrotheria, xenarthra, other mammals, birds & reptiles, amphibians, fishes, other chordates and other eukaryotes. An image from Ensembl was used to show phylogenetics of the 50 animals and an equivalent treeview was built on the left part. Users can browse data by clicking the logos of family and species or by clicking the name on the left treeview of the browse pages. In AnimalTFDB, a cascading style is applied for data browsing, which is browsed by the steps species->families->family gene list->single gene annotation or families->species->family gene list->single gene annotation (Figure 1).

Figure 1.

An overview and gene annotation page in AnimalTFDB. (A) Species in AnimalTFDB. (B) Three kinds of factors in human: TFs, transcription cofactors and chromatin remodeling factors. (C) A list of human TFs in the TF_Otx family. (D) An example of gene annotation page.

Data search

AnimalTFDB provided two different ways to search the data: quick search and advanced search. A quick search box was shown at the top-right of each page designed for searching by Ensembl IDs for gene, transcript and protein, Entrez gene ID or gene symbol. Advanced search page provided multiple ways for searching by different annotations and keywords of each gene. In addition, users could assign the specific families and species for better search.

Comparison with other databases and evaluation of TF identification

We compared our predicted human and mouse TFs with those published by DBD (12) and TFCat (8) databases. DBD is a comprehensive predicted TF database for bacteria, archaea and eukaryotes, while TFCat is a curated catalog for human and mouse TFs. For DBD database, through converting the protein ID into gene ID, we obtained 1383 and 1386 Ensembl gene IDs for human and mouse TF genes, respectively. By comparison, the AnimalTFDB includes 93.7% of human TFs and 93.6% of mouse TFs from DBD database. For the TFs in TFCat database, after ID conversion, we got 521 and 543 Ensembl gene IDs for human and mouse TFs, respectively. The compared result showed that 97.1% of human TFs and 96.3% of mouse TFs from TFCat database were available in our database.

We carefully checked the difference between our AnimalTFDB with the two other databases. For those TFs in the two databases but not in our database, there are two cases. First, some of them are not true TFs predicted by false TF DBD models, such as zf-A20, RNA_pol_Rpb2 and SART-1. Second, some of them should be transcription cofactors or chromatin remodeling factors, which are in the corresponding lists of AnimalTFDB. We also examined the approximately 300 AnimalTFDB-specific TFs for human and mouse. The results showed that some of them were predicted by our unique TF families, such as THAP, CBF, TSC22, Nrf1 and COE. Proteins in these families are true TFs evidenced by publications or having a typical DBD. About half of AnimalTFDB specific TFs were distributed in zf-C2H2, Homeobox, HMG and MYB families, which are all big TF families and account for ∼60% TFs of the genome. Although most of the specific TFs in these big families are unknown proteins containing typical DBDs, we still found a few of them (e.g. KLF6, KLF8, PBX2, TCF7L1 and HBP1) are proved to be as TFs by experiments in publications. Thus, we think we should keep them in the database.

Furthermore, we used the GO annotations to evaluate the reliability and accuracy of our TF list. As a result, we found that 96.3% of our identified human TFs were annotated by TF-related GO terms, such as ‘TF activity’, ‘transcription activator/repressor/regulator activity’ and ‘DNA binding’. These results suggest that the TF prediction approach we used has a reliable performance.

Comparing to other databases, our AnimalTFDB have a more complete and accurate TF family list, and thus a more accurate TF gene list with higher sensitivity and specificity. Moreover, our website is intuitive and easy to browse and search for users. Thirdly, comprehensive annotations are provided in our database as described above. Therefore, we think the AnimalTFDB database will be helpful for the community.