Hotspot Hunter: a computational system for large-scale screening and selection of candidate immunological hotspots in pathogen proteomes
BMC Bioinformatics volume 9, Article number: S19 (2008)
T-cell epitopes that promiscuously bind to multiple alleles of a human leukocyte antigen (HLA) supertype are prime targets for development of vaccines and immunotherapies because they are relevant to a large proportion of the human population. The presence of clusters of promiscuous T-cell epitopes, immunological hotspots, has been observed in several antigens. These clusters may be exploited to facilitate the development of epitope-based vaccines by selecting a small number of hotspots that can elicit all of the required T-cell activation functions. Given the large size of pathogen proteomes, including of variant strains, computational tools are necessary for automated screening and selection of immunological hotspots.
Hotspot Hunter is a web-based computational system for large-scale screening and selection of candidate immunological hotspots in pathogen proteomes through analysis of antigenic diversity. It allows screening and selection of hotspots specific to four common HLA supertypes, namely HLA class I A2, A3, B7 and class II DR. The system uses Artificial Neural Network and Support Vector Machine methods as predictive engines. Soft computing principles were employed to integrate the prediction results produced by both methods for robust prediction performance. Experimental validation of the predictions showed that Hotspot Hunter can successfully identify majority of the real hotspots. Users can predict hotspots from a single protein sequence, or from a set of aligned protein sequences representing pathogen proteome. The latter feature provides a global view of the localizations of the hotspots in the proteome set, enabling analysis of antigenic diversity and shift of hotspots across protein variants. The system also allows the integration of prediction results of the four supertypes for identification of hotspots common across multiple supertypes. The target selection feature of the system shortlists candidate peptide hotspots for the formulation of an epitope-based vaccine that could be effective against multiple variants of the pathogen and applicable to a large proportion of the human population.
Hotspot Hunter is publicly accessible at http://antigen.i2r.a-star.edu.sg/hh/. It is a new generation computational tool aiding in epitope-based vaccine design.
The binding of fragments of processed antigens by major histocompatibility complex (MHC) molecules of antigen presenting cells (APCs) and their presentation to T-cells is crucial for immune surveillance and defence against bacteria, parasites, viruses and tumors. Recognition of the MHC-restricted target peptide on the surface of APCs by the surveying T-cells of the immune system is mediated through the T-cell receptors (TCRs) [1, 2]. Peptides that are recognized by the TCRs and trigger an immune response are called T-cell epitopes and are essential for initiation and regulation of immune responses. Identification of T-cell epitopes in pathogen proteomes is, therefore, crucial for the design of vaccines and immunotherapies. Mapping of these epitopes experimentally, however, is a challenging task because of the large size of pathogens proteomes , great diversity of MHC molecules , and the low (~0.1–5%) natural prevalence of T-cell epitopes for a given MHC molecule . The high cost of peptide synthesis, limited access to human peripheral blood samples, and time-consuming experimental assays further add to the challenge. Experimental approaches are therefore combined with a number of prediction tools to screen for candidate MHC binders, putative T-cell epitopes. This combination has dramatically accelerated the process of epitope mapping as the judicious use of the tools enable large number of laboratory experiments to be avoided .
It can be postulated that an epitope-based vaccine ideally should contain a minimal number of epitopes that cover a vast majority of the human population [7–10]. T-cell epitopes that promiscuously bind to multiple alleles of a human leukocyte antigen (HLA, human MHC) are prime targets for vaccine and immunotherapy development because they are relevant to larger proportions of the human population. The presence of clusters of promiscuous T-cell epitopes, immunological hotspots, has been observed in several antigens, such as SARS coronavirus nucleocapsid , HIV-1 proteins [12–14], and Chlamydia trachomatis outer membrane protein . These clusters may be exploited to facilitate the development of epitope-based vaccines by selecting a small number of hotspots that can elicit all of the required T-cell functions .
Nearly all existing publicly available promiscuous T-cell epitope prediction servers perform prediction for a single protein sequence per submission, while those that accept multiple sequences are not tailor-made to predict hotspots; they rely on manual visualization techniques. Given the large size of pathogen proteomes, it is painstaking to integrate the individual prediction results, and therefore impractical to use these tools for large-scale systematic study of promiscuous epitopes, which is necessary for a global view of the localizations of the epitopes in the proteome and analysis of their antigenic diversity. Herein, we present Hotspot Hunter, a web-based computational system for large-scale screening and selection of candidate immunological hotspots in pathogen proteomes through analysis of antigenic diversity . It allows screening and selection of hotspots specific to four common HLA supertypes, namely HLA class I A2, A3, B7, and class II DR. The A2, A3 and B7 supertypes together cover approximately 88% of the human population [10, 16], irrespective of ethnicity, while class II DR supertype is present in 100% of the population. Hotspot Hunter uses Artificial Neural Network (ANN) and Support Vector Machine (SVM) methods as predictive engines.
It has been reported that combining predictions by several methods results in greater accuracy [17–19]. A soft computing approach was applied to integrate the prediction results produced by both methods used by Hotspot Hunter and the results are presented to users in a succinct and easily understood format. Soft computing is a partnership of several methods, each of them are complementary, not competitive, offering their own advantages to allow solutions to otherwise difficult to solve problems . Importantly, soft computing exploits the tolerance for imprecision, uncertainty and approximation, all characteristics for T-cell epitope data.
The predominant length of peptides that bind HLA class I molecules (HLA-A, -B, and -C) is nine amino acids . HLA class II molecules (HLA-DR) bind longer peptides although through a nine amino acids long binding core [22, 23]. The HLA-A2 training dataset had a total of 3050 (675 binders and 2375 non-binders) 9-mer peptides related to 15 variants of the HLA-A2 supertype (0201, 0202, 0203, 0204, 0205, 0206, 0207, 0208, 0209, 0210, 0211, 0214, 0217, 6802 and 6901). The HLA-A3 training dataset had a total of 2216 (680 binders and 1536 non-binders) 9-mer peptides related to eight variants (0301, 0302, 1101, 1102, 3101, 3301, 3303 and 6801). The HLA-B7 training dataset had a total of 4102 (1258 binders and 2844 non-binders) 9-mer peptides related to 13 variants (0702, 1508, 3501, 3502, 3503, 5101, 5102, 5103, 5301, 5401, 5501, 5502 and 5601). These data were mainly from three sources, the MHCPEP database , IEDB , and published literature, and a set of HLA non-binding peptides (Brusic V., unpublished data). The HLA-A2, -A3 and -B7 datasets are available for download at http://antigen.i2r.a-star.edu.sg/hh/data. The HLA-DR training dataset had 2396 (448 binders and 1948 non-binders) 9-mer peptides related to six HLA-DRB1 variants (0101, 0301, 0701, 0801, 1101 and 1501). The training data was from experiments conducted on 340 15-mers to measure their binding affinity to each of the six HLA-DR alleles. The 15-mers came from three protein sources: MEL40 (aka SSX2, NCBI Accession: AAH16957.1, human melanoma cancer), SSP2 (malaria sporozoite surface protein 2, NCBI Accession: Q01443, Plasmodium yoelli), HCV (Hepatitis C virus genome polyprotein, NCBI Accession: P26663). The transformation from 15-mers to 9-mers was performed utilizing SYFPEITHI motifs . Each 15-mer binder was decomposed into overlapping 9-mers which were submitted to SYFPEITHI for HLA-DR binding prediction. The highest scoring 9-mer was considered as a binder and the rest of the 9-mers were not used in the final 9-mer dataset. On the other hand, all overlapping 9-mers decomposed from 15-mer non-binders were considered as non-binders. Since the data originated from a single set of experiments, with relatively consistent experimental conditions, it is reasonable to expect that the DR training dataset is of higher quality relative to datasets aggregated from multiple sources.
The ANN and SVM models employed herein were the same as those used in MULTIPRED1 . Three-layer back propagation networks (267-4-1) with sigmoid activation functions were built for HLA-A2 and -A3 supertype . Training parameters were determined by observing 100 cross-validation runs. The maximum number of the ANN training cycles was set to 300. The values of momentum and learning rate were 0.5 and 0.001, respectively. The training was repeated four times, and four sets of weights were obtained. The final prediction score was the average of the four predictions calculated using the four sets of weights. Three-layer back propagation networks (289-8-1) with sigmoid activation functions were built for HLA-B7 supertype. The maximum number of the ANN training cycles was set to 500. The training was repeated four times, and four sets of weights were obtained. The values of momentum and learning rate were 0.5 and 0.005, respectively. The HLA-DR supertype was built in a similar fashion by a 4-layer back propagation network (268-2-4-1) with a hyperbolic tangent sigmoid activation function between the two hidden layers and a sigmoid activation function between the second hidden layer and the output. The value of momentum was 0.9, whereas the learning rate was not fixed but changed according to the learning process to achieve faster convergence. The initial learning rate was 0.004. In each training epoch, the sum of square errors was compared with that of the previous epoch. If the sum of square errors increased to more than 1.005 times of the previous one, the learning rate was decreased to 0.7 time of the initial learning rate. If the sum of square errors decreased to less than 0.98 times of the previous one, the learning rate was increased to 1.05 time of the initial learning rate. Ten-fold cross-validation results showed that the area under the receiver operating curve (Aroc) for A2, A3, B7 and DR ANN models were 0.83, 0.83, 0.88 and 0.85, respectively.
A SVM with Gaussian kernel (g = 0.1 and c = 0.5) was used for HLA-A2 and Gaussian kernel (g = 0.1 and c = 2) for HLA-A3 supertype . SVMs with two-degree polynomial kernel function were employed for prediction of peptide binding to HLA-B7 and -DR supertype. Ten-fold cross-validation results showed that the Aroc of A2, A3, B7 and DR SVM models are 0.91, 0.95, 0.92, and 0.80 respectively. The selection of architectures, parameter values, activation functions and kernel functions was done through sampling and comparison of performance.
The prediction performances of models for HLA-A2 and -A3 supertypes were validated using experimental results from a systematic study of human papillomavirus type 16 E6 (NCBI Accession: P03126) and E7 (NCBI Accession: P03129) proteins . Two of three HLA-A2 hotspots and all three HLA-A3 hotspots were correctly predicted without false positive [29, 31]. Recently, Peters et al.  conducted a benchmark study to compare the performance of various bioinformatics models in predicting MHC class I binding peptides and reported that our ANN models  showed the best performance among external tools in predicting peptides binding to three HLA-A2 supertype alleles (0202, 0203 and 0206) and two HLA-A3 supertype alleles (0301 and 1101). Our SVM models have also been validated using the Peters' datasets. The HLA-A2 SVM model outperformed all the external tools evaluated in  and the HLA-A3 SVM model was equal to or better than three of the five HLA-A3 external tools studied . The prediction performances of models for HLA-B7 supertype were validated using experimental results of the tumor-associated antigen survivin (NCBI Accession: NP_001159.2) [34, 35]. Hotspot Hunter correctly predicted one hotspot and missed one. The prediction performance for HLA-DR hotspots was validated using experimental results from systematic binding studies of overlapping peptides from myelin oligodendrocyte glycoprotein (MOG) (NCBI Accession: CAA88109)  and hepatitis C virus 1B protein (NCBI Accession: AAB00216) . All the predicted hotspots were localized in the experimentally validated hotspot regions. Additional validation study using secretory aspartyl proteinase 2 (Sap2), a major protein which is known to induce immune response during Candida infection in human, showed that Hotspot Hunter accurately identified two of the three HLA-DR restricted immunological regions experimentally identified using PBMC proliferation and IL-2 ELISpot assays .
Each input protein sequence to Hotspot Hunter is truncated into overlapping 9-mer peptide sequences with an 8 amino acid overlap and analysis is carried out on each individual 9-mer. To predict immunological hotspots (regions of high concentration of 9-mer promiscuous binders), we have developed two scoring schemes for HLA class I and class II supertypes. The scheme for HLA class I supertypes is based on high scoring individual 9-mers within a window of 30 amino acids [31, 33], and the scheme for HLA class II supertype is based on average scores of individual 9-mers within a window of 15 amino acids. The selection of window length was based on a trial-and-error based heuristics. Window lengths of 15, 20, 25 and 30 were explored and the results were compared with the representative experimental results . The window length of 30 amino acids was found to be best for class I predictions and the window length of 15 amino acids was best for class II predictions. A region in protein sequence is considered as immunological hotspot specific to a HLA supertype only if its predicted binding strength is above the threshold defined for the supertype. The threshold for each supertype was the sum of the thresholds of corresponding ANN and SVM models, which were selected based on experimental binding data [29, 33]. Two flowcharts (Figure S1 and S2) describing the steps involved in identifying hotspots in a single input protein are provided at http://antigen.i2r.a-star.edu.sg/hh/HTML/faq.html#Q3.
The web interface of Hotspot Hunter is intuitive, making use of a set of simple Graphical User Interface forms. Programs were built using a combination of Perl, CGI and C language. The implementation was carried out in SunOS 5.9 UNIX environment.
Using the system
Analysis of antigenic regions in a single protein is initiated by opening the "Single Sequence Query" page of Hotspot Hunter, the query protein sequence is pasted into the query box, a name is assigned and the HLA supertype of interest is selected. The input protein sequence should be in FASTA format and at least 15 (for HLA-DR supertype) or 30 (for HLA-A2/A3/B7 supertype) amino acids long and at most 2000 amino acids long. Input sequences that contain symbols other than amino acids (spaces and carriage returns are allowed) or are of lengths outside the acceptable length range, are not be processed – an error message will be displayed. The sequence pasted in the query box is treated as one single protein sequence (carriage returns are ignored). Figure 1 shows the analysis of the human myelin-oligodendrocyte glycoprotein (MOG) for HLA-DR hotspots. The input interface is shown in Fig 1A (numbering is relative to the mature sequence of MOG). Experimental HLA-DR restricted peptides in the mature sequence of MOG (NCBI Accession: CAA88109) formed two hotspots, 73–96 and 121–216 . Hotspot Hunter predicted four hotspots, 38–56, 79–93, 121–163 and 174–211, for the protein (Figure 1B), three of which corresponded to the two experimentally identified regions. The "binding strength" represents the sum of binding scores predicted by ANN and SVM models for each hotspot. The predicted hotspots are displayed in two formats: in ascending order of their positions in the sequence and in descending order of their prediction scores. Users can click on "Plot Binding Strength" on the result page to generate a graph providing a clear overall view of the binding capacity of the input sequence. The X-axis represents the start position of a peptide (window size of 15-amino acids for DR supertype and 30-amino acids for A2, A3 and B7 supertypes) and Y-axis represents the predicted score of the peptide. Figure 1C shows a plot of the binding scores for the MOG protein with each dot (in blue or red color) representing a 15-amino acids peptide. For example, the first 15-amino acid peptide is plotted at position 1 with a score of 63.53. Individual peptides with prediction scores grater than or equal to the threshold 65 for DR supertype are considered as positive binders to multiple DR alleles and are displayed as red dots, while the non-binders below the threshold are displayed as blue dots.
"Multiple Sequences Query" function allows users to predict hotspots in a set of protein sequences by submitting a multiple sequence alignment file either in ALN (ClustalW) or PHY (PHYLIP) format http://www.ebi.ac.uk/help/formats_frame.html. Predictions are performed on each sequence in the alignment and gaps in the sequences are removed prior to the prediction. However, the gaps are still displayed on the results page and, therefore, hotspots can appear on the results page interspersed by gaps of variable sizes, depending on the alignment. The results page provides a global view of the localizations of the hotspots in the proteome set, enabling analysis of antigenic diversity and shift of hotspots across protein variants. An example of the Hotspot Hunter analysis of a set of dengue virus type 1 capsid sequences is shown in Figure 2. The users may bookmark the URL of the result page (shown in Figure 2B) to access the results which are kept for 24 hours on the server. Alternatively, users can input their email address in the textbox (Figure 2A) and the result file will be emailed to them once the prediction process is complete. The consensus of the input protein sequences is displayed on top of the results page (Figure 2C). The predicted hotspot regions in each protein sequence of the set are highlighted in yellow. Consensus hotspots are defined as hotspots present in at least 50% of the sequences (subject to presence in at least two sequences submitted for prediction) and are displayed at the bottom of the result page. These consensus hotspots are of interest to immunologists as they are shared across multiple protein variant sequences.
"Target Selection" function for multiple sequence queries employs a computational method to identify candidate peptides for the formulation of an epitope-based vaccine that could be effective against multiple variants of the pathogen and is also applicable to a large proportion of the human population. The predicted peptides from consensus hotspots of aligned protein sequences are analyzed for selection of best targets for further experimental validation. All the predicted consensus hotspots are ranked to aid in the decision support, aiming to select the best consensus hotspot. Figure S4 on Hotspot Hunter FAQ page http://antigen.i2r.a-star.edu.sg/hh/HTML/faq.html#Q5 summarizes the process of decision-making in selecting the best consensus hotspot and best candidate targets. The output of this analysis is a comprehensive report, which facilitates interpretation of results and the selection of validation experiments. A sample of the target selection analysis output for dengue virus type 1 NS2b protein is provided in Figure 3.
Hotspot Hunter performs predictions on four HLA supertypes, HLA-A2, -A3, -B7 and -DR. It allows integration of the prediction results of the four supertypes for identification of hotspots common across multiple supertypes, which is facilitated by the function "Hotspots Common Across Supertypes". The user needs to upload the output files (generated by "Multiple Sequences Query") of the supertypes of interest for the same set of sequences. Hotspots common across any combination of the four supertypes can be analysed.
Experimental approaches for identification of T-cell epitopes are laborious and costly, and thus are not applicable for large-scale screening across multiple HLA alleles and pathogen proteomes. Several publicly available online computational systems have been developed for the prediction of peptides binding to HLA alleles and supertypes, such as SYFPEITHI  based on binding motif, BIMAS , ProPred1  and PEPVAC  based on quantitative matrices, SMM  based on stabilized matrix method, MHCPred  based on a multivariate statistical method, SVMHC [43, 44] based on SVM. Combined computational methods that integrate multiple critical steps of MHC class I antigen processing pathways, such as proteasome cleavage, TAP (transporter associated protein) transport, and MHC class I binding, have been recently proposed as a supporting methodology for prediction of high probability targets for therapeutic peptides and vaccines . Several combined computational models of antigen processing and presentation have been reported, such as NetCTL  and MHC-PATHWAY . A major weakness of these methods is that they only represent the major antigen processing pathways; alternative pathways for class I peptide loading exist . For example, proteasomal cleavage is a statistical event with preference for, but not exclusive to, certain cleavage sites. In addition, TAP binding is not the only route for peptide loading into the endoplasmic reticulum (ER); some peptides (signal peptides, membrane peptides, and some viral peptides) are able to access the ER in a TAP-independent manner . Moreover, TAP can transport peptides longer than the optimal HLA-binding length, with endoproteases trimming the peptide to their optimal size in the ER. Taken together, these observations suggest that a simple sequential combination of prediction systems is not adequate. Both relevance and adequacy of such combined systems should be taken with caution. These concerns have been addressed in NetCTL where user can provide the numerical weight for each of the steps in antigen processing and presentation pathway.
Hotspot Hunter is different from other online servers in several aspects. First, the prediction engine combines the strengths of the ANN and SVM methods for better and more robust prediction performance; second, it presents users with predicted immunological hotspots, which are regions of high concentration of predicted promiscuous HLA binding peptides; third, it is suitable for systematic studies of a large set of pathogen proteomes as it can concurrently analyze multiple sequences and present a map providing a global view of their localizations in the proteome, which is the main novelty of the system; and finally, the system provides a utility for selecting candidate experimental targets based on antigenic diversity analysis. Therefore, Hotspot Hunter is a new generation computational tool aiding in epitope-based vaccine design. A customized version of Hotspot Hunter has been integrated into some of our in-house specialized databases, such as the Tumor Antigen Database http://research.i2r.a-star.edu.sg/Templar/DB/cancer_antigen/ and CandiVF – Candida albicans Virulence Factor Database , for prediction of T-cell epitope hotspots.
Human Leukocyte Antigen
Major Histocompatibility Complex
Artificial Neural Network
Support Vector Machine
Antigen Presenting Cell
National Center for Biotechnology Information
Myelin Oligodendrocyte Glycoprotein
Immune Epitope Database and Analysis Resource
Transporter Associated Protein
Hepatitis C Virus
Peripheral Blood Mononuclear Cell.
Janeway CAJ, Travers P, Walport M, Capra JD: Immunobiology. London: Elsevier Science; 1999.
Hudson AW, Ploegh HL: The cell biology of antigen presentation. Exp Cell Res 2002, 272: 1–7. 10.1006/excr.2001.5402
Doytchinova IA, Flower DR: Quantitative approaches to computational vaccinology. Immunol Cell Biol 2002, 80(3):270–279. 10.1046/j.1440-1711.2002.01076.x
Williams TM: Human leukocyte antigen gene polymorphism and the histocompatibility laboratory. J Mol Diagn 2001, 3: 98–104.
Brusic V, Zeleznikow J: Computational binding assays of antigenic peptides. Lett Pept Sci 1999, 6: 313–324.
Brusic V, Bajic VB, Petrovsky N: Computational methods for prediction of T-cell epitopes – a framework for modelling, testing, and applications. Methods 2004, 34(4):436–443. 10.1016/j.ymeth.2004.06.006
Buteau C, Markovic SN, Celis E: Challenges in the development of effective peptide vaccines for cancer. Mayo Clin Proc 2002, 77(4):339–349.
Fischer W, Perkins S, Theiler J, Bhattacharya T, Yusim K, Funkhouser R, Kuiken C, Haynes B, Letvin NL, Walker BD, et al.: Polyvalent vaccines for optimal coverage of potential T-cell epitopes in global HIV-1 variants. Nat Med 2007, 13(1):100–106. 10.1038/nm1461
Khan AM, Heiny A, Lee KX, Srinivasan K, Tan TW, August JT, Brusic V: Large-scale analysis of antigenic diversity of T-cell epitopes in dengue virus. BMC Bioinformatics 2006, 7(Suppl 5):S4. 10.1186/1471-2105-7-S5-S4
Reche PA, Reinherz EL: PEPVAC: a web server for multi-epitope vaccine development based on the prediction of supertypic MHC ligands. Nucleic Acids Res 2005, (33 Web Server):W138–142. 10.1093/nar/gki357
Gupta V, Tabiin T, Sun K, Chandrasekaran A, Anwar A, Yang K, Chikhlikar P, Salmon J, Brusic V, Marques E, et al.: SARS coronavirus nucleocapsid immunodominant T-cell epitope cluster is common to both exogenous recombinant and endogenous DNA-encoded immunogens. Virology 2006, 347(1):127–139. 10.1016/j.virol.2005.11.042
Berzofsky JA, Pendleton CD, Clerici M, Ahlers J, Lucey DR, Putney SD, Shearer GM: Construction of peptides encompassing multideterminant clusters of human immunodeficiency virus envelope to induce in vitro T cell responses in mice and humans of multiple MHC types. J Clin Invest 1991, 88(3):876–884.
Surman S, Lockey TD, Slobod KS, Jones B, Riberdy JM, White SW, Doherty PC, Hurwitz JL: Localization of CD4+ T cell epitope hotspots to exposed strands of HIV envelope glycoprotein suggests structural influences on antigen processing. Proc Natl Acad Sci USA 2001, 98(8):4587–4592. 10.1073/pnas.071063898
Brown SA, Stambas J, Zhan X, Slobod KS, Coleclough C, Zirkel A, Surman S, White SW, Doherty PC, Hurwitz JL: Clustering of Th cell epitopes on exposed regions of HIV envelope despite defects in antibody activity. J Immunol 2003, 171(8):4140–4148.
Kim SK, DeMars R: Epitope clusters in the major outer membrane protein of Chlamydia trachomatis. Curr Opin Immunol 2001, 13(4):429–436. 10.1016/S0952-7915(00)00237-5
Sette A, Sidney J: Nine major HLA class I supertypes account for the vast preponderance of HLA-A and -B polymorphism. Immunogenetics 1999, 50(3–4):201–212. 10.1007/s002510050594
Trost B, Bickis M, Kusalik A: Strength in numbers: achieving greater accuracy in MHC-I binding prediction by combining the results from multiple prediction tools. Immunome Res 2007, 3: 5. 10.1186/1745-7580-3-5
Bhasin M, Raghava GP: A hybrid approach for predicting promiscuous MHC class I restricted T cell epitopes. J Biosci 2007, 32(1):31–42. 10.1007/s12038-007-0004-5
King RD, Ouali M, Strong AT, Aly A, Elmaghraby A, Kantardzic M, Page D: Is it better to combine predictions? Protein Eng 2000, 13(1):15–19. 10.1093/protein/13.1.15
Zadeh LA: Fuzzy Logic and Soft computing: Issues, Contention and Perspectives. IIZUKA'94: 3rd International Conference on Fuzzy Logic, Neural Nets and Soft Computing: 1994; Iisuka, Japan 1994, 1–2.
Rammensee HG, Falk K, Rotzschke O: Peptides naturally presented by MHC class I molecules. Annu Rev Immunol 1993, 11: 213–244. 10.1146/annurev.iy.11.040193.001241
Stern L, Brown J, Jardetzky T, Gorga J, Urban R, Strominger J, Wiley D: Crystal structure of the human class II MHC protein HLA-DR1 complexed with an influenza virus peptide. Nature 1994, 368(6468):215–221. 10.1038/368215a0
Rammensee HG: Chemistry of peptides associated with MHC class I and class II molecules. Curr Opin Immunol 1995, 7(1):85–96. 10.1016/0952-7915(95)80033-6
Brusic V, Rudy G, Harrison LC: MHCPEP: a database of MHC-binding peptides. Nucleic Acids Res 1994, 22: 3663–3665. 10.1093/nar/22.17.3663
Peters B, Sidney J, Bourne P, Bui HH, Buus S, Doh G, Fleri W, Kronenberg M, Kubo R, Lund O, et al.: The immune epitope database and analysis resource: from vision to blueprint. PLoS Biol 2005, 3(3):e91. 10.1371/journal.pbio.0030091
Rammensee HG, Bachmann J, Emmerich NP, Bachor OA, Stevanovic S: SYFPEITHI: database for MHC ligands and peptide motifs. Immunogenetics 1999, 50(3–4):213–219. 10.1007/s002510050595
Zhang GL, Kwoh CK, August JT, Brusic V: Performance Evaluation of MULTIPRED1 on Prediction of MHC Class I Binders. International Conference on Biomedical and Pharmaceutical Engineering: 2006; Singapore 2006.
Zhang GL, Khan AM, Srinivasan KN, August JT, Brusic V: Neural Models for predicting viral vaccine targets. J Bioinform Comput Biol 2005, 3(5):1207–1225. 10.1142/S0219720005001466
Zhang GL, Bozic I, Kwoh CK, August JT, Brusic V: Prediction of supertype-specific HLA class I binding peptides using support vector machines. J Immunol Methods 2007, 320: 143–154. 10.1016/j.jim.2006.12.011
Kast WM, Brandt RM, Sidney J, Drijfhout JW, Kubo RT, Grey HM, Melief CJ, Sette A: Role of HLA-A motifs in identification of potential CTL epitopes in human papillomavirus type 16 E6 and E7 proteins. J Immunol 1994, 152: 3904–3912.
Srinivasan KN, Zhang GL, Khan AM, August JT, Brusic V: Prediction of Class I T-cell epitopes: evidence of presence of immunological hot spots inside antigens. Bioinformatics 2004, 20(Suppl 1):i297-i302. 10.1093/bioinformatics/bth943
Peters B, Bui HH, Frankild S, Nielson M, Lundegaard C, Kostem E, Basch D, Lamberth K, Harndahl M, Fleri W, et al.: A community resource benchmarking predictions of peptide binding to MHC-I molecules. PLoS Comput Biol 2006, 2(6):e65. 10.1371/journal.pcbi.0020065
Zhang GL, Khan AM, Srinivasan KN, August JT, Brusic V: MULTIPRED: a computational system for prediction of promiscuous HLA binding peptides. Nucleic Acids Res 2005, 33: W172-W179. 10.1093/nar/gki452
Kessler JH, Mommaas B, Mutis T, Huijbers I, Vissers D, Benckhuijsen WE, Schreuder GM, Offringa R, Goulmy E, Melief CJ, et al.: Competition-based cellular peptide binding assays for 13 prevalent HLA class I alleles using fluorescein-labeled synthetic peptides. Hum Immunol 2003, 64(2):245–255. 10.1016/S0198-8859(02)00787-5
Bachinsky MM, Guillen DE, Patel SR, Singleton J, Chen C, Soltis DA, Tussey LG: Mapping and binding analysis of peptides derived from the tumor-associated antigen survivin for eight HLA alleles. Cancer Immun 2005, 5: 6.
Weissert R, Kuhle J, de Graaf KL, Wienhold W, Herrmann MM, Muller C, Forsthuber TG, Wiesmuller KH, Melms A: High immunogenicity of intracellular myelin oligodendrocyte glycoprotein epitopes. J Immunol 2002, 169(1):548–556.
Borras-Cuesta F, Golvano J, Garcia-Granero M, Sarobe P, Riezu-Boj J, Huarte E, Lasarte J: Specific and general HLA-DR binding motifs: comparison of algorithms. Hum Immunol 2000, 61(3):266–278. 10.1016/S0198-8859(99)00153-6
Tongchusak S: In silico prediction and immunological validation of candida albicans secretory aspartyl proteinase 2 peptides bound to multiple HLA-DRB1 supertypes. PhD thesis. Mahidol University; 2006.
Parker KC, Bednarek MA, Coligan JE: Scheme for ranking potential HLA-A2 binding peptides based on independent binding of individual peptide side-chains. J Immunol 1994, 152(1):163–175.
Singh H, Raghava GP: ProPred1: prediction of promiscuous MHC Class-I binding sites. Bioinformatics 2003, 19(8):1009–1014. 10.1093/bioinformatics/btg108
Peters B, Sette A: Generating quantitative models describing the sequence specificity of biological processes with the stabilized matrix method. BMC Bioinformatics 2005, 6: 132. 10.1186/1471-2105-6-132
Hattotuwagama CK, Guan P, Doytchinova IA, Zygouri C, Flower DR: Quantitative online prediction of peptide binding to the major histocompatibility complex. J Mol Graph Model 2004, 22(3):195–207. 10.1016/S1093-3263(03)00160-8
Donnes P, Kohlbacher O: SVMHC: a server for prediction of MHC-binding peptides. Nucleic Acids Res 2006, (34 Web Server):W194–197. 10.1093/nar/gkl284
Donnes P, Elofsson A: Prediction of MHC class I binding peptides, using SVMHC. BMC Bioinformatics 2002, 3(1):25–38. 10.1186/1471-2105-3-25
Petrovsky N, Brusic V: Virtual models of the HLA class I antigen processing pathway. Methods 2004, 34(4):429–435. 10.1016/j.ymeth.2004.06.005
Larsen M, Lundegaard C, Lamberth K, Buus S, Brunak S, Lund O, Nielsen M: An integrative approach to CTL epitope prediction: a combined algorithm integrating MHC class I binding, TAP transport efficiency, and proteasomal cleavage predictions. Eur J Immunol 2005, 35(8):2295–2303. 10.1002/eji.200425811
Tenzer S, Peters B, Bulik S, Schoor O, Lemmel C, Schatz MM, Kloetzel PM, Rammensee HG, Schild H, Holzhutter HG: Modeling the MHC class I pathway by combining predictions of proteasomal cleavage, TAP transport and MHC class I binding. Cell Mol Life Sci 2005, 62(9):1025–1037. 10.1007/s00018-005-4528-2
Gromme M, Neefjes J: Antigen degradation or presentation by MHC class I molecules via classical and non-classical pathways. Mol Immunol 2002, 39(3–4):181–202. 10.1016/S0161-5890(02)00101-3
Brusic V, van Endert P, Zeleznikow J, Daniel S, Hammer J, Petrovsky N: A neural network model approach to the study of human TAP transporter. In Silico Biol 1999, 1(2):109–121.
Tongchusak S, Chaiyaroj SC, Veeramani A, Koh JLY, Brusic V: CandiVF – Candida albicans virulence factor database. Int J Pep Res Ther 2005, 11(4):271–277. 10.1007/s10989-005-9268-5
The authors thank Seng Hong Seah, Judice Koh and Olivo Miotto for their help and valuable suggestions. This project has been funded in part with Federal funds from the National Institute of Allergy and Infectious Diseases, National Institutes of Health, Department of Health and Human Services, USA, under Grant No. 5 U19 AI56541 and Contract No. HHSN2662-00400085C; and the ImmunoGrid project under the EC contract FP6-2004-IST-4, No. 028069.
This article has been published as part of BMC Bioinformatics Volume 9 Supplement 1, 2008: Asia Pacific Bioinformatics Network (APBioNet) Sixth International Conference on Bioinformatics (InCoB2007). The full contents of the supplement are available online at http://www.biomedcentral.com/1471-2105/9?issue=S1.
The authors declare that they have no competing interests.
GLZ and AMK performed the in silico experiments and drafted the manuscript. KNS, ATH, KXL, JTA and VB participated in the design of the study. VB conceived the study, participated in its design and coordination and helped draft the manuscript. JTA, VB and CKK critically reviewed the manuscript. All authors read and approved the final manuscript.
Guang Lan Zhang, Asif M Khan contributed equally to this work.
About this article
Cite this article
Zhang, G.L., Khan, A.M., Srinivasan, K.N. et al. Hotspot Hunter: a computational system for large-scale screening and selection of candidate immunological hotspots in pathogen proteomes. BMC Bioinformatics 9 (Suppl 1), S19 (2008). https://doi.org/10.1186/1471-2105-9-S1-S19
- Human Leukocyte Antigen
- Support Vector Machine Model
- Human Leukocyte Antigen Class
- Myelin Oligodendrocyte Glycoprotein
- NCBI Accession