Background

Protein Post-Translational Modification (PTM) plays a critical role in cellular control mechanism, including phosphorylation for signal transduction, attachment of fatty acids for membrane anchoring and association, glycosylation for changing protein half-life, targeting substrates, and promoting cell-cell and cell-matrix interactions, and acetylation and methylation of histone for gene regulation [1]. Several databases collecting information about protein modifications have been established through high-throughput mass spectrometry in proteomics. UniProtKB/Swiss-Prot [2] collects many protein modification information with annotation and structure. Phospho.ELM [3], PhosphoSite [4] and Phosphorylation Site Database [5] were developed for accumulating experimentally verified phosphorylation sites. PHOSIDA [6] integrates thousands of high-confidence in vivo phosphorylation sites identified by mass spectrometry-based proteomics in various species. Phospho 3D [7] is a database of 3D structures of phosphorylation sites, which stores information retrieved from the phospho.ELM database and is enriched with structural information and annotations at the residue level. O-GLYCBASE [8] is a database of glycoproteins, most of which include experimentally verified O-linked glycosylation sites. UbiProt [9] stores experimental ubiquitylated proteins and ubiquitylation sites, which are implicated in protein degradation through an intracellular ATP-dependent proteolytic system. Moreover, the RESID protein modification database is a comprehensive collection of annotations and structures for protein modifications and cross-links, including pre-, co-, and post-translational modifications [10].

dbPTM [11] was developed previously to integrate several databases to accumulate known protein modifications, as well as the putative protein modifications predicted by a series of accurately computational tools [12, 13]. This updated version of dbPTM was enhanced to become a knowledge base for protein post-translational modifications, which comprises a variety of new features including the modified sites, solvent accessibility of substrate, protein secondary and tertiary structures, protein domains and protein variations. We also collected literature related to PTM, protein conservations and the specificity of substrate site. Especially for protein phosphorylation, the site-specific interactions between catalytic kinases and substrates are provided. Furthermore, a variety of prediction tools have been developed for more than ten PTM types [14], such as phosphorylation, glycosylation, acetylation, methylation, sulfation and sumoylation. This work constructed a benchmark data set for computational studies of protein post-translational modification. The benchmark data set can provide a standard for measuring the performance of prediction tools that have been presented for identifying post-translational modification sites of proteins. The web interface of dbPTM is also redesigned and enhanced to facilitate the access to the proposed resource.

Data construction and content

As shown in Figure 1, the system architecture of dbPTM2.0 database comprises three major components: the integration of external PTM databases, the computational identification of PTMs, and the structural and functional annotations of PTMs. We integrated five PTM databases, including UniProtKB/Swiss-Prot (release 55.0) [1], Phospho.ELM (version 7.0) [15], O-GLYCBASE (version 6.0) [8], UbiProt (version 1.0) [9] and PHOSIDA (version 1.0) [6] for obtaining experimental protein modifications. The description and data statistics of these databases are briefly given in Table S1 (see Additional file 1 – Table S1). Additionally, Human Protein Reference Database (HPRD) [16], which compiles invaluable information relevant to functions and PTMs of human proteins in health and disease, was also integrated.

Figure 1
figure 1

The system architecture of the knowledge base for protein translational modification. It comprises the three major components: integration of external experimental PTM databases, learning and prediction of 20 types of PTM, and annotations of PTM knowledge (more details in the text).

Full size image

In the part of computational identification of PTMs, KinasePhos-like method [11–13, 17] was applied for identifying 20 types of PTM, which contain at least 30 experimentally verified PTM sites. The detailed processing flow of KinasePhos-like methods is displayed in Figure S1 (See Additional file 1 – Figure S1). The learned models were evaluated using k-fold cross validation. Table S2 (See Additional file 1 – Table S2) lists the predictive performance of these models. To reduce the number of false positive predictions, the predictive parameters were set to ensure a maximal of predictive specificity.

The statistics of the experimental PTM sites and putative PTM sites in this integral PTM database is given in Table 1. After removing the redundant PTM sites among six databases, there are totally 45833 experimental PTM sites in this update version. All experimental PTM sites are further categorized by PTM types. For instance, there are 31, 363 experimental phosphorylation sites and 2,080 experimental acetylation sites in the database. In addition to the experimental PTM sites, UniProtKB/Swiss-Prot provides putative PTM sites by using sequence similarity or evolutionary potential. Moreover, KinasePhos-like methods [11–13, 17] were adopted to construct the profile hidden Markov models (HMMs) for twenty types of PTMs. These models were applied to identify the potential PTM sites against protein sequences obtained from UniProtKB/Swiss-Prot. As given in Table 1, 2,560,047 sites for all PTM types were identified. The structural and functional annotations of protein modifications were obtained from UniProtKB/Swiss-Prot [18], InterPro [19], Protein Data Bank [14]. To understand the predictive performance of these tools previously developed, it is crucial to have a common standard for evaluating the predictive performance among various prediction tools. Therefore, we constructed a benchmark, which comprise the experimental substrate sequences for each PTM type.

The process to compile the evaluation sets is described in Figure S3 (See Additional file 1 – Figure S3), based on criteria developed by Chen et al. [30]. To remove the redundancy, the protein sequences containing the same type of PTM sites are grouped by a threshold of 30% identity by BLASTCLUST [31]. If the identity of two protein sequences is greater than 30%, we re-aligned the fragment sequences of the substrates by BL2SEQ. If the fragment sequences of two substrates with the same location are identical, only one of the substrate was included in the benchmark data set. Therefore, twenty PTM types containing more than 30 experimental sites were complied in the benchmark data set.

Enhanced web interface

A user-friendly web interface is provided for simple searching, browsing, and downloading of protein PTM data. In addition to the database query by the protein name, gene name, UniProtKB/Swiss-Prot ID or accession, it allows the input of protein sequences for similarity search against UniProtKB/Swiss-Prot protein sequences (See Additional file 1 – Figure S4). To provide an overview of PTM types and their modified residues, a summary table is provided for browsing the information and the annotations about the post-translational modification types, which are referred to the UniProtKB/Swiss-Prot PTM list http://www.expasy.org/cgi-bin/lists?ptmlist.txt and RESID [10].

Figure 3 shows an example that users can choose the acetylation of lysine (K) to obtain more detailed information such as the position of modified amino acid, the location of the modification in protein sequence, the modified chemical formula, the mass difference, and the substrate site specificity, which is the preference of amino acids surrounding the modification sites. Furthermore, the structural information, such as solvent accessibility and secondary structure surrounding the modified sites, are provided. All the experimental PTM sites and putative PTM sites can be downloaded from the web interface.

Figure 3
figure 3

An illustrative example to show the catalytic specificity of acetyllysine.

Full size image

Conclusion

The proposed server enables both wet-lab biologists and bioinformatics researchers to easily explore the information about protein post-translational modifications. This study not only accumulates the experimentally verified PTM sites with relevant literature references, but also computationally annotates twenty types of PTM sites against UniProtKB/Swiss-Prot proteins. As given in Table 2, the proposed knowledge base provides effective information of protein PTMs, including sequence conservation, subcellular localization and substrate specificity, the average solvent accessibility and the secondary structure surrounding the modified site. Moreover, we construct a PTM benchmark data set that can be adopted for computational studies in evaluating the predictive performance of various tools about determining PTM sites. Previous investigations have indicated that many protein modifications cause binding domains for specific protein-protein interaction to regulate cellular behavior [32]. All the experimental PTM sites and putative PTM sites are available and downloadable in the web interface. Prospective work of dbPTM is to integrate protein-protein interaction data.

Availability and requirements

Project name: dbPTM 2.0: A Knowledge Base for Protein Post-Translational Modifications

ASMD project home page: http://dbPTM.mbc.nctu.edu.tw/

Operating system(s): Platform-independent

Programming Language: PHP, Perl

Other requirements: a modern web browser (with CSS and JavaScript support)

Restrictions to use by non-academics: None