
==== Front
Nucleic Acids ResNucleic Acids ResnarnarNucleic Acids Research0305-10481362-4962Oxford University Press 1755383310.1093/nar/gkm390ArticlesDOMAC: an accurate, hybrid protein domain prediction server Cheng Jianlin *School of Electrical Engineering and Computer Science, University of Central Florida, Orlando, FL, 32816, USA*To whom correspondence should be addressed. (407) 823-0230(407) 823-5419jcheng@cs.ucf.edu7 2007 6 6 2007 6 6 2007 35 Web Server issue W354 W356 12 1 2007 28 3 2007 1 5 2007 © 2007 The Author(s)2007This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/2.0/uk/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.Protein domain prediction is important for protein structure prediction, structure determination, function annotation, mutagenesis analysis and protein engineering. Here we describe an accurate protein domain prediction server (DOMAC) combining both template-based and ab initio methods. The preliminary version of the server was ranked among the top domain prediction servers in the seventh edition of Critical Assessment of Techniques for Protein Structure Prediction (CASP7), 2006. DOMAC server and datasets are available at: http://www.bioinfotool.org/domac.html
==== Body
INTRODUCTION
Protein domains are structural, functional and evolutionary units of proteins. The prediction of domains from sequence information can improve tertiary structure prediction (1), enhance protein function annotation (2), aid structure determination (3) and guide protein engineering (4) and mutagenesis (5).

A number of different methods have been developed to identify domains starting from primary sequences. These methods can be roughly classified into four categories: template-based methods (6–10), ab initio (template-free) methods (11–22), the hybrid approach combining template-based and ab initio methods (23), and meta-domain prediction methods (24).

Here we describe an accurate, hybrid domain prediction server (DOMAC) that integrates homology modeling, domain parsing and ab initio methods together. The preliminary implementation of the server [under the name: FOLDpro (25)] participated in the domain evaluation in the seventh edition of Critical Assessment of Techniques for Protein Structure Prediction (CASP7) (26,27). It was ranked among the top domain prediction servers in CASP7.

IMPLEMENTATION
Our hybrid approach uses the template-based method to predict domains for proteins having homologous template structures in Protein Data Bank (PDB) (28), and the ab initio method based on neural networks (29) to predict domains for de novo proteins. It predicts protein domains in two steps.

First, it uses the PSI-BLAST (30) to search the target sequence against NCBI Non-Redundant sequence database to construct a profile. The profile is used to search a template structure library built from the proteins in PDB to identify templates, similarly as PDB-BLAST approach (31).

Second, if some significant templates are identified (e-value ⩽0.001), it generates a structure model for the target using Modeller (32) based on the template structures. Multiple significant templates are combined to improve model quality if available. Then it uses an accurate domain parsing tool PDP (33) to parse the model into domains. If the parsed domains do not cover the whole target sequence, DOMAC will assign uncovered regions to adjacent domains.

If no significant homologous template is found, DOMAC will invoke the ab initio domain predictor DOMpro (29) to predict domains. DOMpro uses neural networks in conjunction with sequence profile, predicted secondary structure, and relative solvent accessibility to predict domain boundary. The secondary structure and relative solvent accessibility are predicted by SSpro (34) and ACCpro (35) in the SCRATCH suite (36). DOMpro tries to identify domain boundary positions based on the composition bias of sequence and structural features in domain linker regions.

The preliminary implementation of DOMAC participated in CASP7 and was ranked first among 13 domain prediction servers. Since then, we have significantly speeded up the template identification process without sacrificing accuracy and added a module to update the template library weekly to incorporate the newly released proteins in PDB.

RESULTS
Here we firstly describe the performance of the preliminary implementation of DOMAC in CASP7 (under server name: FOLDpro). We compare it with 12 other server predictors in CASP7 using two evaluation metrics: CASP evaluation metric (37) and domain number accuracy.

CASP metric (NDO: normalized domain overlap score) is to compute the overlapping score of domains without explicitly checking domain number and domain boundary (37). It computes the numbers of correctly and wrongly overlapped residues between true domains and predicted domains, respectively. It summarizes the numbers of the overlapping residues into a single score to evaluate domain prediction. The best score for a target is 1 and the worst score is 0. The domain number accuracy is defined as the percentage of targets with correct domain number predictions.

Table 1 reports the performance of 13 servers on 95 targets in CASP 7. The CASP score is the average domain overlap score across all predicted targets. The domain number accuracy is computed by comparing the domain number predictions with the official domain definitions released by CASP7. In terms of the two evaluation metrics, the preliminary implementation of DOMAC (FOLDpro) yielded the best performance.
Table 1. The performance of 13 domain prediction servers in CASP7

Method	Target Num	Domain Num Acc. (%)	CASP7 Score	
FOLDpro (DOMAC)	95	93.7	0.963	
Baker-RosettaDom (23)	94	86.2	0.940	
Ma-OPUS-DOM	94	87.2	0.933	
ROBETTA-GINZU (23)	94	84.0	0.932	
DomSSEA (7)	94	78.7	0.910	
HHpred3 (38)	95	75.8	0.910	
Meta-DP (24)	95	74.7	0.907	
HHpred1 (38)	93	75.3	0.902	
DomFOLD	95	75.8	0.898	
DPS(13)	93	75.3	0.889	
Chop (22)	83	56.6	0.827	
Distill (39)	95	70.5	0.819	
NN_PUT-Lab	92	58.7	0.795	
The second column (target num) lists the number of targets for which a predictor made predictions.



We also evaluate DOMAC on the three categories of CASP7 targets: highly homologous, homologous and analogous/ab initio. The domain number prediction accuracy of DOMAC is 96%, 94% and 88% in the three categories, respectively.

However, because the majority (68 out of 95) of CASP7 targets is single-domain proteins, the domain prediction accuracy is very likely over-estimated.

Thus, we evaluate DOMAC on a larger, balanced, high-quality dataset manually curated by Holland et al. (2). The publicly released version of the Holland's benchmark2 dataset has 156 proteins consisting of 54 single-domain proteins, 69 two-domain proteins, 25 three-domain proteins, 4 four-domain proteins, 3 five-domain proteins and 1 six-domain protein. We evaluate both template-based and ab initio methods on the whole dataset, respectively. Table 2 reports the specificity and sensitivity of each method in each category in terms of domain numbers. The overall domain number prediction accuracy of the template-based and ab initio methods is 75% and 46%, respectively.
Table 2. The specificity and sensitivity of domain number prediction on the Holland's dataset using the template-based and ab initio methods

Method	Acc. (%)	1-dom	2-dom	3-dom	4-dom	5-dom	6-dom	
Template	Sens.	96.1	66.7	56.0	75.0	66.7	–	
	Spec.	74.2	88.0	70.0	42.9	33.3	–	
Ab initio	Sens.	88.5	31.3	12.0	–	–	–	
	Spec.	46.5	48.8	30.0	–	–	–	


Moreover, we assess the accuracy of the domain boundary prediction, which is important for generating hypotheses for crystallizing individual protein domains. Following the same convention (7,22), a predicted boundary within 20 residues away from a true domain boundary is considered correct.

The domain boundary specificity and sensitivity is 50% and 76.5% for the template-based method, and 27% and 14% for the ab initio method. Thus, the accuracy are sufficient for guiding the crystallization experiment, whereas the ab initio method is not always reliable enough for the general, practical use.

USE OF WEB SERVICE
The use of DOMAC are intuitive through a simple input form. Since the reliability assessment of domain predictions is still an open issue, the user is advised to use the accuracy on the Holland's dataset to decide how to use these predictions. The input form requires only three inputs: email address, target name, and protein sequence. DOMAC usually can make predictions within 15 min and send the results back to users through email.

Domain prediction results include the user-defined target name, the protein sequence, the predicted domain number, the start and end positions of each domain and the method (template-based or ab initio). For template-based prediction, it also reports the PDB codes of the templates. Figure 1 shows an output example for the CASP7 target T0324.
Figure 1. Domain prediction result of CASP7 target T0324. The protein is predicted to have two domains. Domain 1 has two non-continuous segments, spanning from residues 1 to 16 and residues 82 to 208, respectively. Domain 2 spans from residues 17 to 81. The templates used to make the domain prediction are identified by PDB code + chain id. The chain in a single-chain protein is always assigned chain id ‘A’ instead of ‘-’.



CONCLUSION AND FUTURE WORK
We have developed a hybrid domain prediction web service integrating template-based and ab initio methods. The template-based method is accurate enough for guiding protein structure prediction, structure determination, function annotation, mutagenesis analysis and protein engineering. However, the ab initio method still needs to be improved for practical use. Since protein domain architecture is largely shaped by gene recombination events, such as gene fusion, fission, domain swapping and exon exchange, leveraging the evolutionary gene recombination signals embedded in the multiple sequence alignment of a protein family and exon boundaries (or splicing sites) in its gene structure, may help improve ab initio domain prediction significantly.

ACKNOWLEDGEMENTS
J.C. is very grateful to Dr Pierre Baldi for the support during his PhD research at University of California Irvine.

Funding to pay the Open Access publication charges for this article was provided by the New Faculty Start-Up Grant at the University of Central Florida.

Conflict of interest statement. None declared.
==== Refs
REFERENCES
1 Chivian D  Kim DE  Malmstrom  L. Bradley P  Robertson T  Murphy P  Strauss CE  Bonneau R  Rohl CA    Automated prediction of CASP-5 structures using the Robetta server Proteins 2003 53 S6 524 533 14579342 
2 Holland T  Veretnik S  Shindyalov IN  Bourne PE   A benchmark for domain assignment from protein 3-dimensional structure and it's applications J. Mol. Biol 2006 361 562 590 16863650 
3 Campbell I  Downing A   Building protein structure and function from modular units Trends Biotechnol 1994 12 168 172 7764899 
4 Guerois R  Serrano L   Protein design based on folding models Curr. Opin. Struct. Biol 2001 11 101 106 11179898 
5 Nielsen P  Yamada Y   Identification of cell-binding sites on the Laminin a5 n-terminal domain by site-directed mutagenesis J. Biol. Chem 2001 276 10906 10912 11098055 
6 Heger A  Holm L   Exhaustive enumeration of protein domain families J. Mol. Biol 2003 328 749 767 12706730 
7 Marsden RL  McGuffin LJ  Jones DT   Rapid protein domain assignment from amino acid sequence using predicted secondary structure Protein Sci 2002 11 2814 2824 12441380 
8 von Ohsen N  Sommer I  Zimmer R  Lengauer T   Arby: automatic protein structure prediction using profile-profile alignment and confidence measures Bioinformatics 2004 20 2228 2235 15059818 
9 Gewehr JE  Zimmer R   SSEP-Domain: protein domain prediction by alignment of secondary structure elements and profiles Bioinformatics 2006 22 181 187 16267083 
10 Coin L  Bateman A  Durbin R   Enhanced protein domain discovery using taxonomy BMC Bioinformat 2004 5 56 
11 Park J  Teichmann SA   DIVCLUS: an automatic method in the GEANFAMMER package that finds homologous domains in single- and multi-domain proteins Bioinformatics 1998 14 144 150 9545446 
12 Gouzy J  Corpet F  Kahn D   Whole genome protein domain analysis using a new method for domain clustering Comput. chem 1999 23 333 340 10404623 
13 Bryson K  McGuffin LJ  Marsden RL  Ward JJ  Sodhi JS  Jones DT   Protein structure prediction servers at University College London Nucleic Acids Res 2005 33 w36 w38 15980489 
14 George RA  Heringa J   SnapDRAGON: a method to delineate protein structural domains from sequence data J. Mol. Biol 2002 316 839 851 11866536 
15 Linding R  Russell RB  Neduva V  Gibson TJ   GlobPlot: Exploring protein sequences for globularity and disorder Nucleic Acids Res 2003 31 3701 3708 12824398 
16 Nagarajan N  Yona G   Automatic prediction of protein domains from sequence information using a hybrid learning system Bioinformatics 2004 20 1335 1360 14962932 
17 Wheelan SJ  Marchler-Bauer A  Bryant SH   Domain size distributions can predict domain boundaries Bioinformatics 2000 16 613 618 11038331 
18 Sim J  Kim SY  Lee J   PPRODO: prediction of protein domain boundaries using neural networks Proteins 2005 59 627 632 15789433 
19 Chen L  Wang W  Ling S  Jia C  Wang F   Kemadom: a web server for domain prediction using kernel machine with local context Nucleic Acids Res 2006 34 W158 w163 16844982 
20 Adams R  Das S  Smith T   Multiple domain protein diagnostic patterns Prot. Sci 1996 5 1240 1249 
21 George R  Heringa J   Protein domain identification and improved sequence similarity search using PSI-BLAST Protein Struct. Funct. Genet 2002 48 672 681 
22 Liu J  Rost B   Sequence-based prediction of protein domains Nucleic Acids Res 2004 32 3522 3530 15240828 
23 Kim DE  Chivian D  Malmstrom L  Baker D   Automated prediction of domain boundaries in casp6 targets using Ginzu and RosettaDOM Proteins 2005 61 Suppl. 7 193 200 16187362 
24 Saini HK  Fischer D   Meta-DP: domain prediction meta server Bioinformatics 2005 21 2917 2920 15840708 
25 Cheng J  Baldi P   A Machine Learning Information Retrieval Approach to Protein Fold Recognition Bioinformatics 2006 22 1456 1463 16547073 
26 Moult J  Fidelis K  Zemla A  Hubbard T   Critical assessment of methods of protein structure prediction (CASP)-round v Proteins 2003 53 Suppl. 6 334 339 14579322 
27 Moult J  Fidelis K  Rost B  Hubbard T  Tramontano A   Critical assessmentof methods of protein structure prediction (CASP) - round 6 Proteins 2005 61 Suppl 7 3 7 16187341 
28 Berman HM  Westbrook J  Feng Z  Gilliland G  Bhat TN  Weissig H  Shindyalov IN  Bourne PE   The Protein Data Bank Nucleic Acids Res 2000 28 235 242 10592235 
29 Cheng J  Sweredoski MJ  Baldi P   DOMpro: Protein domain prediction using profiles, secondary structure, relative solvent accessibility, and recursive neural networks Data Mining and Knowledge Discovery 2006 13 1 10 
30 Altschul SF  Madden TL  Schaffer AA  Zhang J  Zhang Z  Miller AA  Lipman DJ   Gapped BLAST and PSI-BLAST: a new generation of protein database search programs Nucleic Acids Res 1997 25 3389 3402 9254694 
31 Rychlewski L  Jaroszewski L  LI W  Godzik A   Comparison of sequence profiles. Strategies for structural predictions using sequence information” Protein Sci 2000 9 232 241 10716175 
32 Sali A  Blundell TL   Comparative protein modelling by satisfaction of spatial restraints J. Mol. Biol 1993 234 779 815 8254673 
33 Alexandrov N  Shindyalov I   PDP: protein domain parser Bioinformatics 2003 19 429 430 12584135 
34 Pollastri G  Przybylski D  Rost B  Baldi P   Improving the prediction of protein secondary structure in three and eight classes using recurrent neural networks and profiles Proteins 2002 47 228 235 11933069 
35 Pollastri G  Baldi P  Fariselli P  Casadio R   Prediction of coordination number and relative solvent accessibility in proteins Proteins 2002 47 142 153 11933061 
36 Cheng J  Randall AZ  Sweredoski MJ  Baldi P   SCRATCH: a protein structure and structural feature prediction server Nucleic Acids Res 2005 33 web server issue w72 w76 15980571 
37 Tai CH  Lee  W.J. Vincent JJ  Lee B   Evaluation of domain prediction in CASP6 Proteins 2005 61 Suppl. 7 183 192 16187361 
38 Söding J   Protein homology detection by HMM-HMM comparison Bioinformatics 2005 21 951 960 15531603 
39 Bau D  Martin AJM  Mooney C  Vullo A  Walsh I  Pollastri G   Distill: A suite of web servers for the prediction of one-, two- and three dimensional structural features of proteins BMC Bioinformat 2006 7 402

