Abstract
Computational prediction of T-cell epitope candidates is currently being used in several applications including vaccine discovery studies, development of diagnostics and removal of unwanted immune responses against protein therapeutics. There have been continuous improvements on the performance of MHC binding prediction tools but their general adoption by immunologists has been slow due to the lack of user-friendly interfaces and guidelines. Current tools only provide minimal advice on what alleles to include, what lengths to consider, how to deal with homologous peptides and what cutoffs should be considered relevant. This protocol provides step-by-step instructions with necessary recommendations for prediction of the best T-cell epitope candidates in line with the newly developed online tool called TepiTool. The TepiTool, part of IEDB, provides some of the top MHC binding prediction algorithms for number of species including humans, chimpanzees, bovines, gorillas, macaques, mice and pigs. The TepiTool is freely accessible at http://tools.iedb.org/tepitool/.
Keywords: MHC class I, MHC class II, T cell epitope, binding affinity prediction, CTL epitope prediction
Introduction
The binding of a peptide to an MHC molecule is necessary for its ability to activate T cell responses. Peptides bind MHC molecules in the “peptide binding groove”, forming a peptide-MHC complex which in turn is recognized by the T cell receptors. Peptides recognized by T cells are called epitopes (Murphy, 2011). Epitopes bound to class I and class II MHC molecules are recognized by CD8+ and CD4+ T cells, respectively. Generally, MHC binding prediction tools scan amino acid sequences to estimate the binding affinity of each component peptide to a specific MHC. MHC class I molecules have a binding groove that is closed at its ends, limiting the size of its ligands to roughly 8-11 residues in length. Class II molecules, on the other hand, have an open binding groove, allowing them to bind longer peptides, typically 12-20 residues in length. The strength of binding (affinity) of a peptide to an MHC molecule is an important factor that determines potential immunogenicity (Sette et al., 1994).
Computational prediction of T cell epitope candidates has been used in many epitope identification and vaccine discovery studies in recent years (De Groot & Berzofsky, 2004; Larsen et al., 2010; Lin, Zhang, Tongchusak, Reinherz, & Brusic, 2008; Lund et al., 2011; Lundegaard, Hoof, Lund, & Nielsen, 2010; Moise et al., 2009; Moutaftsi et al., 2006; Sette & Rappuoli, 2010; Sylvester-Hvid et al., 2004). Predictions are also helpful in development of diagnostics (He et al., 2004; Vincenti et al., 2003; M. Wang et al., 2007) and identification and removal of unwanted immune responses against protein-based drugs or consumer products (Oseroff et al., 2010; Paul et al., 2013a; Tangri et al., 2005). Indeed, pre-screening peptides on the basis of predicted MHC binding affinity can save valuable time and resources needed for the epitope identification process. Different methods of computational binding prediction algorithms have been developed, including those based on artificial neural networks (Buus et al., 2003), support vector machines (Bhasin & Raghava, 2004), matrix based prediction methods in which the matrix coefficients are fitted in a machine learning process (Kim, Sidney, Pinilla, Sette, & Peters, 2009), and matrix based predictions in which the coefficients are directly measured based on positional scanning of combinatorial peptide libraries (PSCL) (Sidney et al., 2008). A major innovation was the introduction of pan-specific MHC binding predictors, in which a single neural network simultaneously learns how to make predictions for all MHC molecules of a given class, thereby enabling binding predictions for MHC alleles for which no specific binding data is available (Hoof et al., 2009). Different computational tools for prediction of binding affinity of peptides to MHC class I and class II molecules are provided at several publically accessible websites, including the Immune Epitope Database (IEDB) (Kim et al., 2012; Salimi, Fleri, Peters, & Sette, 2010), BIMAS (Parker, Bednarek, & Coligan, 1994), SYFPEITHI (Rammensee, Bachmann, Emmerich, Bachor, & Stevanović, 1999), NetMHC (Lundegaard et al., 2008), ProPred (Singh & Raghava, 2001), ProPred1 (Singh & Raghava, 2003), MULTIPRED (G. L. Zhang, Khan, Srinivasan, August, & Brusic, 2005) and Rankpep (Reche & Reinherz, 2007). The IEDB's analysis resource (http://tools.immuneepitope.org/main/) (Q. Zhang et al., 2008) currently hosts T cell epitope prediction tools for MHC class I and class II alleles that provide some of the best prediction algorithms for hundreds of alleles (Moutaftsi et al., 2006; Trolle et al., 2015; P. Wang et al., 2008; P. Wang et al., 2010).
While there have been continuous improvements on the algorithmic performance of MHC binding predictions, a hurdle for their general adoption by immunologists has been the lack of user friendly interfaces and guidelines. Current tools focus on returning scores corresponding to binding affinities for a set of peptides. As maintainers of the IEDB Analysis resource, we are frequently asked how such predictions should be utilized in practice, such as for a study of T cell responses against a given pathogen in the human population. Current tools only provide minimal advice on what alleles to include, what lengths to consider, how to deal with homologous peptides, and what cutoffs should be considered relevant.
To address this, we have developed a set of step-by-step instructions on how to perform predictions for frequently occurring use cases. We are documenting these guidelines in this paper, and have in parallel designed a new online tool called TepiTool, for computational prediction of T-cell epitope candidates based on peptide binding to MHC class I and class II alleles. The tool is freely accessible to the public at http://tools.iedb.org/tepitool/. The TepiTool has implemented some of the top MHC binding prediction algorithms available for hundreds of alleles from different species, including humans, chimpanzees, bovines, gorillas, macaques, mice and pigs. This unit describes prediction of T cell epitope candidates using the TepiTool.
Basic Protocol
Computational Prediction of Peptides Binding to MHC Class I AND Class II Molecules
This protocol explains prediction of T cell epitope candidates from a given set of amino acid sequences, based on predicted peptide binding to MHC class I and class II molecules, using the online computational MHC binding prediction tool called TepiTool. The tool is designed as a wizard where the user is led through a series of well-defined steps to complete the task. Each step is a client-side web form that takes user input data that is in turn processed at the server-side when the user submits the entire form. All fields except sequences and alleles are filled with default recommended settings for prediction and selection of optimum peptides. The input parameters can be adjusted as per the user's specific needs, and the user can go back to previous steps to change the selection before final submission of the job. The TepiTool has six steps as described below.
Materials
Computer with internet browser and proper internet connection
Protein sequence(s) for binding prediction in single letter amino acid code.
Protocol steps
The TepiTool has six steps as listed below.
Provide sequence data
Select the host species and MHC allele class
Select the alleles for binding prediction
Select peptides to be included in prediction
Select preferred methods for binding prediction and peptide selection and cutoff values
Review selections, enter job details and submit data
Step-1: Provide sequence data
The first step in the prediction process is to provide the sequences that should be scanned for MHC binding peptides. All sequences have to be amino acids specified in single letter code (ACDEFGHIKLMNPQRSTVWY) in FASTA or plain text format. The sequences can either be entered directly into the text area provided or uploaded as a text file (Figure-1). In FASTA format a sequence begins with a single-line description, followed by lines of sequence data. The description line should start with a greater than symbol (“>”). Example of FASTA sequence format is given in supplementary file 1. If “>” symbol is not found, the sequences will be considered to be in PLAIN format and each line will be considered a separate sequence.
Figure 1.
Step-1 in the MHC binding prediction using TepiTool. The figure shows the form for entering sequences to be scanned for MHC binding peptides.
Step-2: Select the host species and MHC allele class
In the second step the user selects the species and MHC class of alleles for which binding predictions will be performed (Figure-2). The user can choose from human, chimpanzee, cow, gorilla, macaque, mouse and pig. For human and mouse, the user can choose between MHC class I and class II. For all other species, only class I predictions are available at this time, as the class II molecules for those species have not been sufficiently characterized. The list of available species and alleles will grow in the future as more data is made available and algorithms are developed. The TepiTool will be updated to provide binding predictions for the new species and alleles. The choices for MHC class in this step correspondingly limit the options available in the Steps 3 - 5.
Figure 2.
The figure shows step-2 where the user can choose the species and MHC allele class for which binding prediction needs to be performed.
Step-3: Select the alleles for binding prediction
In the third step the user needs to specify the alleles from the species chosen for binding prediction. Depending on the choice in the preceding step, alleles are chosen for class I (Step-3a) or class II (Step-3b).
Step-3a: MHC class I allele selection
For all species the user has the option of either choosing the alleles from the list of available alleles provided or upload a list of alleles in the specified format as a plain text file. When alleles are uploaded as a file, alleles should be listed as one allele per row and the allele names should follow the proper nomenclature. Sample format for human is given below. Sample formats for other species are given in supplementary file 2, and the IMGT website may also be consulted (http://www.imgt.org/) (Lefranc et al., 1999).
HLA-A*01:01
HLA-A*01:02
HLA-A*01:03
For more details on nomenclature of HLA alleles please refer HLA nomenclature website (http://hla.alleles.org) (Robinson et al., 2015). For information on nomenclature of MHC alleles from other species please refer the Immuno Polymorphism Database (IPD-MHC, http://www.ebi.ac.uk/ipd/mhc/) (Ellis et al., 2006). For humans additional options to choose alleles are available (Figure-3a), as explained below. The options available include:
Figure 3a.
This shows step-3 of the TepiTool where the user can select the MHC class I alleles that need to be predicted for.
Select from a list of frequently occurring alleles: This list provides the alleles with frequency > 1% in the global population (http://www.allelefrequencies.net/) (Gonzalez-Galarza et al., 2015).
Select from list of all available alleles: This list provides all alleles for which the IEDB has binding prediction algorithms available.
Select from list of representative alleles from different HLA supertypes: This list provides a set of representative alleles from 12 different MHC class I supertypes (Sidney, Peters, Frahm, Brander, & Sette, 2008).
Select panel of 27 most frequent alleles: This list contains 27 most frequent alleles in the global population (Weiskopf et al., 2013).
Step-3b: MHC class II allele selection
Binding predictions for MHC class II is available for only human and mouse. For mouse, the options available are either choosing the alleles from the list of available alleles provided or upload a list of alleles in the specified format as a text file. There are different options available for choosing alleles for human MHC class II, as described below (Figure-3b).
Figure 3b.
This shows step-3 of the TepiTool where the user can select the MHC class II alleles from the list of available alleles provided.
-
Predict for custom allele set, where the user can select desired alleles from a list of available alleles. Separate lists are provided for DP, DQ and DR loci. The default option shows alleles that are frequent in the global population. By clicking the checkbox provided the user can choose α and β chains separately for DP and DQ alleles (this option is not available for DR, since only the DR-β chain is effectively variable). The user can also upload the list of alleles as a plain text file similar to class I. Alleles should be listed as one allele per row. Sample format for humans is given below. The format for mouse alleles is given in supplementary file 2.
HLA-DPA1*01:03/DPB1*02:01
HLA-DQA1*01:02/DQB1*06:02
HLA-DRB1*01:01
HLA-DRB1*01:02
Predict for pre-selected panel of alleles: This option currently provides a set of 26 most frequent human class II alleles from DP, DQ and DR loci (Figure-3c). The user has the option to choose alleles from each locus separately or any combination of the three loci.
-
Predict using pre-selected allele sets and methods: This option predicts a peptide's ability to bind to multiple alleles (i.e., “promiscuous binding”) in pre-defined sets of alleles (Figure-3d). There are two options for predictions based on promiscuous binding:
The “7-allele method”: This option does prediction based on the median of the consensus percentile ranks for 7 common DR alleles (HLA-DRB1*03:01, HLA-DRB1*07:01, HLA-DRB1*15:01, HLA-DRB3*01:01, HLA-DRB3*02:02, HLA-DRB4*01:01, HLA-DRB5*01:01). Peptides with median consensus percentile rank ≤ 20.0 are selected as predicted binders (this cutoff can be adjusted in step 6). The selection of these alleles and the establishment of this cutoff were based on a broad evaluation of epitope prediction efficacy using class II binding predictions in (Paul et al., 2015).
Promiscuous binding prediction using the 26 most frequent and/or representative alleles: In this option, the prediction is done based on the number of alleles in a set of 26 most frequent in the global population (Greenbaum et al., 2011) that each peptide is predicted to bind. A consensus percentile rank ≤ 20.0 is used as cutoff to define binding. As a default, peptides predicted to bind ≥ 50% of the alleles are selected as promiscuous binders. This approach precedes the 7-allele-method, and was based on evaluation of prediction performance in a single dataset in (Oseroff et al., 2010).
Figure 3c.
This shows step-3 of the TepiTool where the user can select the “pre-selected” panel of 26 most frequent MHC class II alleles for prediction.
Figure 3d.
This shows step-3 of the TepiTool where the user can select the “pre-selected allele sets and methods” for prediction.
Step-4: Select peptides to be included in prediction
In this step the user has the option to define parameters to provide an approximate number and nature of peptides to be included in the prediction. That is, specific parameters will allow for the definition of large sets of peptides based on more permissive prediction criteria, or limited sets based on more stringent criteria.
Step-4a: Peptide selection for MHC class I alleles
The first field, “Select peptides to be included in prediction”, provides the option to choose either default settings for low/moderate/high number of peptides or custom select the peptides to be included in the prediction which in turn determines the number of peptides in the prediction results (Figure-4a). In case of option “low number of peptides”, only 9mer peptides are included and duplicate peptides are removed. When the option “moderate number of peptides” is chosen 8mers, 9mers, 10mers and 11mers are included and duplicate peptides are removed. The option “high number of peptides” includes all possible peptides from 8mers to 14mers and retains duplicate peptides. If “custom selection” is chosen the user can adjust the handling of duplicate peptides and select peptide lengths as desired (Figure-4b).
Figure 4a.
This shows step-4 which provides different options for selecting the number and nature of peptides to be included in MHC class I binding prediction.
Figure 4b.
This shows step-4 where the user can custom select peptides to be included in MHC class I binding prediction.
The option, “handling of duplicate peptides”, allows the user to remove or retain identical peptides from different sequences in the result. That is, if a peptide sequence meeting the input parameters is present in more than one input sequence, the peptide will be included in the results only once if the “remove duplicate peptides” option is chosen, whereas it will be included as many times if “keep duplicate peptides” option is chosen instead. Retaining duplicate peptides may be desirable in certain occasions because they generate homogenous “stand-alone” peptide sets for analysis or testing.
The option “peptide lengths to be considered for prediction” provides the choices to select the peptide lengths to be included in the prediction. The user has the option to choose from 8mers to 14mers. This field also shows the approximate number of peptides that will be included in the prediction process against each length.
The second field, “conservancy analysis”, allows the user to select only the peptides that are conserved in a specified % of the input sequences. This is useful especially in case of homologous sequences where the user wants to pick only the most conserved peptides, such as when the user wants to predict epitope candidates for a protein for which many isoforms exist. The user also has the option to specify the desired % of sequences as cutoff if conservancy analysis is to be done.
Step-4b: Peptides selection for MHC class II alleles
The options available for MHC class II are mostly the same as for MHC class I. One major difference is that in case of MHC class II the peptide length is fixed at 15. MHC class II molecules have an open binding groove (compared to the closed binding groove of MHC class I molecules) and can accommodate longer peptides compared to class I (Castellino, Zhong, & Germain, 1997; Chicz et al., 1992; Rudensky, Preston-Hurlburt, Hong, Barlow, & Janeway, 1991; Stern et al., 1994). However, the majority of the energy of the interaction between a peptide and a class II molecule is provided by a peptide core of about 9 amino acids in length. The presence of additional amino acids flanking the binding core also appears to be necessary for stable binding, although they do not specifically interact with the MHC peptide-binding groove. Accordingly, in common practice peptides of 15 residues are used for MHC class II binding predictions. Selecting 15mer peptides overlapping by 10 amino acid residues will provide the minimum number of peptides with all possible 9mer binding cores that has at least one flanking residue on both sides. In this way the possibility of selecting redundant peptides with the same binding core can be avoided.
Similar to class I, the first field, “Select peptides to be included in prediction”, provides the option of choosing either default settings for low/moderate/high number of peptides or custom select the peptides to be included in prediction (Figure-4c). In case of option “low number of predicted peptides”, the 15mers generated are overlapping by 8 AA residues and duplicate peptides are removed. The option “moderate number of peptides” increases the number of peptides included by increasing the number of overlapping residues to 10. The duplicate peptides are removed here as well. When the option “high number of peptides” is chosen, the number of overlapping residues remains same at 10, but duplicate peptides are retained. If “custom selection” is chosen the user can adjust the handling of duplicate peptides and no. of overlapping residues as desired (Figure-4d).
Figure 4c.
This shows step-4 which provides different options for selecting the number and nature of peptides to be included in MHC class II binding prediction.
Figure 4d.
This shows step-4 where the user can custom select peptides to be included in MHC class II binding prediction.
Similar to class I, the option, “handling of duplicate peptides”, allows the user to remove or retain identical peptides from different sequences in the results.
The option, “desired number of overlapping residues”, provides the option to choose the number of overlapping amino acid residues for the 15mers to be generated. The default is set for 10 residues, but the user has the option to choose any number between 0-14 residues.
The last section of the field shows the approximate no. of peptides to be considered for prediction based on the selections for handling of duplicate peptides and number of overlapping residues.
Similar to class I, the last field, “conservancy analysis”, allows the user to select only the peptides that are conserved in a specified % of input sequences.
The above options are available only if the user selects “predict for custom allele set” or “predict for pre-selected panel of alleles” option in step-3. If the user selects “predict pre-selected allele sets and methods” in step-3, the duplicate peptides will always be removed and number of overlapping residues will be fixed at 10.
Step-5: Select preferred methods for binding prediction and peptide selection and cutoff values
Step-5a: Select preferred methods for binding prediction and peptide selection and cutoff values for MHC class I alleles
Here the first field provides the user the option to choose a specific binding prediction method (Figure-5). The prediction method list box allows choosing from a number of MHC class I binding prediction methods, including IEDB recommended, Consensus (Moutaftsi et al., 2006), NetMHCpan (version 2.8) (Hoof et al., 2009; Nielsen et al., 2007), ANN (Artificial neural network, also called as NetMHC, version 3.4) (Buus et al., 2003; Lundegaard, Nielsen, & Lund, 2006; Lundegaard et al., 2008; Lundegaard, Lund, & Nielsen, 2008; Nielsen et al., 2003), SMM (Stabilized matrix method) (Peters & Sette, 2005), SMMPMBEC (SMM with a Peptide:MHC Binding Energy Covariance matrix)(Kim et al., 2009) and Comblib_Sidney2008 (scoring matrices derived from PSCL)(Sidney et al., 2008).
Figure 5.
This figure shows step-5 which provides the options to choose the binding prediction method and the strategy to select the predicted peptides.
“IEDB recommended” is the default prediction method. Based on availability of predictors and previously observed predictive performance, this selection tries to use the best possible method for a given MHC molecule. Currently for peptide:MHC-I binding predictions, for a given MHC molecule the IEDB recommended method represents a consensus across the ANN, SMM, and CombLib predictors, or a subset thereof, depending on the availability of the respective methods for the specific allele probed. Otherwise, NetMHCpan is used. This choice was motivated by the expected predictive performance of the methods in decreasing order: Consensus > ANN > SMM > NetMHCpan > CombLib (Moutaftsi et al., 2006). This order may be updated in future depending on the results from future benchmarking studies (Trolle et al., 2015).
It should be noted that we fully expect the IEDB recommendation to change as we perform larger benchmarks of newly developed methods on blind datasets to determine an accurate assessment of prediction quality. For that purpose, automated benchmarks have been established to continuously evaluate the performance of the existing MHC class I binding methods. These benchmarks are updated weekly as new datasets are deposited into the IEDB and the latest results can be found at http://tools.immuneepitope.org/auto_bench/mhci/ (Trolle et al., 2015).
The second field provides options of different approaches for selection of predicted peptides. There are five different ways you can choose the predicted peptides.
Select peptides based on predicted percentile rank: In this approach, the peptides will be selected based on the predicted percentile rank. A percentile rank is generated by comparing the peptide's predicted binding affinity against a large set of randomly selected peptides from the Swiss-Prot database. Percentile scores provide a uniform scale allowing comparisons across different predictors. A lower percentile rank indicates higher affinity. In the case of the consensus method, the median percentile rank of the three methods involved is used to generate the consensus percentile rank. The user has the option to provide the cutoff percentile rank for determining binders. The default value provided is 1.0, i.e. all peptides with percentile rank ≤ 1.0 will be selected as predicted peptides.
Select peptides based on predicted IC50: Here the peptide selection is based on the predicted IC50 nM. The user has the option to provide the cutoff IC50 value for determining binders. The default value provided is 500 nM, i.e. all peptides with predicted IC50 ≤ 500 nM will be selected as predicted peptides. Please note that if this approach is chosen, the prediction method will always be NetMHCpan, irrespective of the prediction method chosen in the first field, as the quantitative predictions of affinity provide by the NetMHCPan method are more uniform across alleles.
Select peptides based on MHC-specific predicted binding threshold: This method is applicable to human only. In this approach the peptides are selected based on allele-specific binding thresholds, as described in Paul et al., 2013 (Paul et al., 2013b). The threshold for each allele is based on previous analyses of set of known T cell epitopes, and corresponds to the expected threshold needed identify the set of peptides accounting for 75% of the total immune response. Please note that this option is available for only a certain number of HLA alleles, as listed in supplementary file 3.
Select top x% of predicted peptides: This option allows the user to simply select the specified % of peptides out of all peptides included in the predictions. For example, if the user wants to select the top 5% 9mers, and there are approximately 100 9mers in total included in the prediction set, this approach will select the top five 9mers as predicted peptides. The peptide selection criterion in this approach will always be predicted percentile rank.
Select top x number of predicted peptides: This option is similar to 4; the difference being that here the user can input the exact number of peptides needed to be selected in the final results instead of %. Here also the peptide selection criterion is always predicted percentile rank.
Step-5b: Select preferred methods for binding prediction and peptide selection and cutoff values for MHC class II alleles
Similar to MHC class I, the first field provides the user the option to choose the binding prediction method. The prediction method list box allows choosing from a number of MHC class II binding prediction methods, namely IEDB recommended, Consensus (P. Wang et al., 2008; P. Wang et al., 2010), NetMHCIIpan (version 3.0) (Karosiene et al., 2013; Nielsen et al., 2008), NN_align (Artificial neural network, also called as NetMHCII-2.2, version 2.2) (Nielsen & Lund, 2009), SMM_align (also called as NetMHCII-1.1, version 1.1) (Nielsen, Lundegaard, & Lund, 2007), Sturniolo (Sturniolo et al., 1999) and Combinatorial library (Scoring matrices derived from combinatorial peptide libraries) .
As above, “IEDB recommended” is the default prediction method selection. Similar to class I, this selection chooses the best possible method for a given MHC class II molecule based on evaluations by the IEDB team and bioinformatics community. Currently for peptide:MHC-II binding predictions, for a given MHC molecule, the IEDB recommended approach uses the Consensus method, consisting of NN_align, SMM_align and Sturniolo/Combinatorial Library, if any corresponding predictor is available for the molecule. Otherwise, if none of therse approaches is available, NetMHCIIpan is used. The Consensus approach considers a combination of any three of the four methods, if available, where Sturniolo is a final choice. The expected predictive performances are based on large scale evaluations of the performance of the MHC class II binding predictions: a 2008 study based on over 10,000 binding affinities, a 2010 study based on over 40,000 binding affinities (P. Wang et al., 2010) and a study comparing pan-specific methods (L. Zhang, Udaka, Mamitsuka, & Zhu, 2012). We expect the IEDB recommendation to change as we perform larger benchmarks on blind datasets to determine an accurate assessment of prediction quality of newly developed or improved methods.
Similar to class I, the second field provides options of different approaches for selection of predicted peptides. There are five different ways to select the predicted peptides:
Select peptides based on predicted percentile rank: In this approach, the peptides will be selected based on the predicted percentile rank. The user has the option to provide the cutoff percentile rank for determining binders. The default value provided is 10.0 which is the recommended cutoff for human MHC class II alleles, i.e. all peptides with percentile rank ≤ 10.0 will be selected as predicted peptides.
Select peptides based on predicted IC50: Here the peptide selection is based on the predicted IC50 nM. The user has the option to provide the cutoff percentile rank for determining binders. The default value provided is 1000 nM, i.e. all peptides with predicted IC50 ≤ 1000 nM will be selected as predicted peptides. Please note that if this approach is chosen, the prediction method will always be NetMHCIIpan irrespective of the prediction method chosen in the first filed.
Select peptides based on the number of alleles bound: This approach is based on the peptide's ability to bind to multiple alleles (referred to as “promiscuous binding”) and is similar to the second option provided in step-3, except that here the users can select their own set of alleles. Percentile rank ≤ 20.0 is used as cutoff for binding to the allele and 50% alleles is chosen by default as the cutoff for number of alleles binding to determine binders. The cutoff here is set as 20.0 because this method involves multiple alleles and too stringent of a cutoff might be overlooking potential promiscuous restrictions. The user has the option to adjust the latter cutoff.
Select top x% of predicted peptides: This option allows the user to select the specified % of peptides out of all peptides included in prediction. For example, if the user wants to select the top 5% 15mers and there are approximately 100 15mers in total that are included in the prediction, this approach will select the top five 15mers as predicted peptides. The peptide selection criterion in this approach will always be predicted percentile rank.
Select top x number of predicted peptides: This option is similar to 4; the difference being that here the user can input the exact number of peptides needed to be selected in the final results instead of %. Here also the peptide selection criterion is always predicted percentile rank.
The above options are available only when “predict for specific allele(s)” option is chosen in step-3. If the option “predict promiscuous binding” is chosen in step-3, the prediction method will always be “IEDB recommended”. If the “7-allele method” is chosen under the “promiscuous binding”, the selection of predicted peptides is based on the median of the consensus percentile rank of the 7 alleles included in this method. All peptides with median consensus percentile rank ≤ 20.0 will be selected by default. This cutoff value can be adjusted by the user. If the promiscuous binding based on the “26 most frequent alleles” is chosen, the peptide selection is based on the number of alleles bound, i.e. peptides that bind to at least 50% of alleles (13 alleles) will be selected.
Step-6: Review selections, enter job details and submit data
In the last step, the user can review the summary of input parameters that were provided (Figure-6). The user also has the option to provide a job name and email id. Providing the job name will make it easy for tool support to troubleshoot submissions, if necessary. If email id is provided, an email will be sent with the concise results once the job is finished. If any of the selections need to be changed, the user can go back to previous steps and make necessary edits at this stage. It should be noted that once the form is submitted the user will not be able to make any more changes. The user will have to start the prediction process all over again if changes are needed. The browser window should not be closed or refreshed until the results are printed.
Figure 6.
This figure shows the last step in the binding prediction process where the user can review the input parameters and enter job name and email address.
Commentary
Critical Parameters and Troubleshooting
The TepiTool should run trouble free as long as proper internet connection is available and valid inputs are provided for binding prediction. If the input parameters provided are not valid the tool will give an error message and will not allow the user to proceed until the error is fixed. Sometimes the prediction process may not fetch any valid results even with proper input parameters. In most cases this may be due to the restrictions placed by input parameters. Relaxing cutoff values may help solve this issue. When a prediction method other than “IEDB recommended” is chosen, it is possible that the selected allele may not have the selected algorithm available and may fail to perform the prediction. In such cases, repeating the prediction with method as “IEDB recommended” may solve it. If “IEDB recommended” method also did not fetch any results, it is probable that predictions are not possible for the particular combination selected (for example, peptide length and allele in case of class I, or the specific combination of α and β chains selected in case of class II). In any case, we can examine the issue in detail if the user contacts the IEDB with the details of the prediction task (at help@iedb.org). Once the job is submitted the browser window should not be closed until the results are obtained.
Anticipated Results
Once the prediction job is submitted at the end of step-6 and the prediction process is finished, the results are displayed on the results page. The result page has the following sections:
1. Concise results
The concise results table shows the final list of predicted peptides selected based on the input parameters provided (Figure-7a). The table will contain the index of the peptide's source protein in the input sequence set, start and end positions of the peptide within the source protein sequence, the peptide sequence, the selection criterion parameter value depending on the chosen peptide selection strategy (percentile rank/IC50/median consensus percentile rank/number of binding alleles), the allele (where applicable) and conservancy % (if conservancy analysis is done, the concise result will also show the conservancy of each peptide within the input sequence set). The concise results table can also be downloaded as csv file which can be opened using any spreadsheet, such as MS Excel, for further analysis. This section will be included in the email sent to the user.
Figure 7a.
This figure shows the concise results section of the results page. The concise results table shows the sequence number of the source protein of the respective peptide, start and end coordinates of the peptide within the source protein, peptide sequence, allele and conservancy % if conservancy analysis is done. The link to download this table as a csv file is also provided.
2. Download results details
Links for downloading the results details as csv files (Figure-7b) are provided here. It can include the following based on the input parameters:
Figure 7b.
This figure shows the section from the results page for downloading the detailed results. Non-redundant results contain prediction results with redundant peptides within each sequence removed. Complete results contain binding predictions of all peptides. Conservancy of peptides contains conservancy of each peptide in the input sequence set.
Non-redundant results: This is applicable only to class II binding predictions. This file will contain prediction results with redundant peptides within each sequence removed. Redundant peptides mean peptides that overlap by a larger number of residues than desired. This result set includes both positive and negative peptides based on the input parameters.
Complete results: This file will contain binding predictions of all peptides. This will include the predicted IC50 and percentile rank or other scores depending on the prediction method chosen. In the case of the IEDB recommended or consensus method, the results will include details from each of the prediction methods employed.
Conservancy of peptides: This is applicable only if conservancy analysis is done. This file will contain conservancy of each peptide in the input sequence set.
3. Citation information
This section provides reference information for the prediction methods utilized, and that should be cited if the prediction results are used in a particular study or manuscript (Figure-7c). This section will also be included in the email sent to the user with prediction concise results.
Figure 7c.
This shows the citation information section from the results page.
4. Input sequences
The input sequence set, along with the sequence description (in case of FASTA format) that was provided for binding prediction (Figure-7d), is provided in this section. In cases where the plain sequence format was utilized, and hence no sequence description was provided, the sequence description/title will be automatically designated in the format of “>Seq_1” etc.
Figure 7d.
This section from the results page shows the sequences that were used in the binding prediction.
5. Other input parameters
All input parameters that were provided for the binding prediction, except the sequences, are indicated in this last section (Figure-7e).
Figure 7e.
This figure shows the section from the results page that shows all input parameters except sequences that were provided for binding prediction.
Interpretation of results
The concise results show the selected peptides based on the binding predictions as per the input parameters provided. In cases of peptide selection based on percentile rank, median consensus percentile rank (7-allele method) and IC50, a lower numerical value indicates better predicted binding affinity. In cases of promiscuous binding based on the number of alleles predicted to be bound, better candidates correspond to those with a high number of alleles predicted. The recommended input parameters for prediction and selection of optimum peptides are provided as default input parameters.
It should be noted that the predicted peptides are only a list of probable epitope candidates, and only experimental verification can be considered decisive.
Time Considerations
Different factors can affect the extent of time required for prediction of binding peptides. Most important among them are: (1) the number of sequences entered for binding prediction and their length - as the number of sequences and their length increase, the amount of computational time increases proportionately; (2) the prediction method selected - the “pan specific” methods such as NetMHCpan and NetMHCIIpan tend to take a longer time to finish, compared to other methods; (3) the number of alleles and their type - time required increases as more alleles are selected for prediction, and alleles for which only “pan specific” methods are available will also take longer time; (4) the number of different peptide lengths (8mers, 9mers etc) in case of class I - in case of class I, choosing more peptide lengths will also increase the amount of time required to complete the prediction process. In general, however, computation time is reasonably quick, and results expected in many cases in less than a minute, but in most cases less than five minutes, but also depending on internet connection speed. As an example, prediction for three sequences (∼1000 amino acids in total) for 5 HLA class I alleles for 9mers and 10mers with method as “IEDB recommended” will be finished in less than 30 seconds in most settings.
Supplementary Material
Significance Statement.
Computational prediction of T-cell epitope candidates is currently being used in several applications including vaccine discovery studies, development of diagnostics and removal of unwanted immune responses against protein therapeutics. This protocol provides step-by-step instructions with necessary recommendations for prediction of the best T-cell epitope candidates in line with the newly developed online tool called TepiTool. The TepiTool provides some of the top MHC class I and class II binding prediction algorithms for number of species including humans, chimpanzees, bovines, gorillas, macaques, mice and pigs. The tool is designed as a user-friendly wizard with well-defined steps which helps the users to predict the best MHC binding peptides from their sequences of interest.
Acknowledgments
This project has been funded with federal funds from the National Institute of Allergy and Infectious Diseases, National Institutes of Health, U.S. Department of Health and Human Services under Contract Number HHSN272201200010C (IEDB). We want to thank Dorjee Tamang, Jivan Amara and Dr. Jason Greenbaum for their valuable inputs during the development of TepiTool.
Footnotes
Internet Resources: 1. TepiTool: The newly developed online tool described in this protocol, for prediction of peptides binding to MHC class I and class II molecules (http://tools.iedb.org/tepitool/).
2. IEDB's analysis resource (http://tools.iedb.org/): A collection of tools for the prediction and analysis of immune epitopes.
3. International ImMunoGeneTics Information system (IMGT, http://www.imgt.org/): An integrated knowledge resource for the immunoglobulins or antibodies, T cell receptors, major histocompatibility of human and other vertebrate species.
4. HLA nomenclature website (http://hla.alleles.org): Website describing HLA allele nomenclature.
5. Immuno Polymorphism Database (IPD-MHC, http://www.ebi.ac.uk/ipd/mhc/): A centralized repository for sequences of the Major Histocompatibility Complex (MHC) from a number of different species and information on their nomenclature.
6. Allelefrequencies.net database (http://www.allelefrequencies.net/): Database with frequencies of HLA alleles.
Contributor Information
Sinu Paul, Email: spaul@lji.org.
John Sidney, Email: jsidney@lji.org.
Alessandro Sette, Email: alex@lji.org.
Bjoern Peters, Email: bpeters@lji.org.
Literature Cited
- Bhasin M, Raghava GP. SVM based method for predicting HLA-DRB1*0401 binding peptides in an antigen sequence. Bioinformatics (Oxford, England) 2004;20(3):421–423. doi: 10.1093/bioinformatics/btg424. [doi] [DOI] [PubMed] [Google Scholar]
- Buus S, Lauemøller S, Worning P, Kesmir C, Frimurer T, Corbet S, Fomsgaard A, Hilden J, Holm A, Brunak S. Sensitive quantitative predictions of peptide-MHC binding by a ‘Query by Committee’artificial neural network approach. Tissue Antigens. 2003;62(5):378–384. doi: 10.1034/j.1399-0039.2003.00112.x. [DOI] [PubMed] [Google Scholar]
- Castellino F, Zhong G, Germain RN. Antigen presentation by MHC class II molecules: Invariant chain function, protein trafficking, and the molecular basis of diverse determinant capture. Human Immunology. 1997;54(2):159–169. doi: 10.1016/s0198-8859(97)00078-5. [DOI] [PubMed] [Google Scholar]
- Chicz RM, Urban RG, Lane WS, Gorga JC, Stern LJ, Vignali DA, Strominger JL. Predominant naturally processed peptides bound to HLA-DR1 are derived from MHC-related molecules and are heterogeneous in size. Nature a-z Index. 1992;358(6389):764–768. doi: 10.1038/358764a0. [DOI] [PubMed] [Google Scholar]
- De Groot AS, Berzofsky JA. From genome to vaccine--new immunoinformatics tools for vaccine design. Methods (San Diego, Calif) 2004;34(4):425–428. doi: 10.1016/j.ymeth.2004.06.004. [DOI] [PubMed] [Google Scholar]
- Ellis SA, Bontrop RE, Antczak DF, Ballingall K, Davies CJ, Kaufman J, Kennedy LJ, Robinson J, Smith DM, Stear MJ. ISAG/IUIS-VIC comparative MHC nomenclature committee report, 2005. Immunogenetics. 2006;57(12):953–958. doi: 10.1007/s00251-005-0071-4. [DOI] [PubMed] [Google Scholar]
- Gonzalez-Galarza FF, Takeshita LY, Santos EJ, Kempson F, Maia MH, da Silva AL, Teles e Silva AL, Ghattaoraya GS, Alfirevic A, Jones AR, Middleton D. Allele frequency net 2015 update: New features for HLA epitopes, KIR and disease and HLA adverse drug reaction associations. Nucleic Acids Research. 2015;43(Database issue):D784–8. doi: 10.1093/nar/gku1166. [doi] [DOI] [PMC free article] [PubMed] [Google Scholar]
- Greenbaum J, Sidney J, Chung J, Brander C, Peters B, Sette A. Functional classification of class II human leukocyte antigen (HLA) molecules reveals seven different supertypes and a surprising degree of repertoire sharing across supertypes. Immunogenetics. 2011;63(6):325–335. doi: 10.1007/s00251-011-0513-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- He Y, Zhou Y, Wu H, Luo B, Chen J, Li W, Jiang S. Identification of immunodominant sites on the spike protein of severe acute respiratory syndrome (SARS) coronavirus: Implication for developing SARS diagnostics and vaccines. Journal of Immunology (Baltimore, Md: 1950) 2004;173(6):4050–4057. doi: 10.4049/jimmunol.173.6.4050. doi:173/6/4050. [pii] [DOI] [PubMed] [Google Scholar]
- Hoof I, Peters B, Sidney J, Pedersen LE, Sette A, Lund O, Buus S, Nielsen M. NetMHCpan, a method for MHC class I binding prediction beyond humans. Immunogenetics. 2009;61(1):1–13. doi: 10.1007/s00251-008-0341-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Karosiene E, Rasmussen M, Blicher T, Lund O, Buus S, Nielsen M. NetMHCIIpan-3.0, a common pan-specific MHC class II prediction method including all three human MHC class II isotypes, HLA-DR, HLA-DP and HLA-DQ. Immunogenetics. 2013;65(10):711–724. doi: 10.1007/s00251-013-0720-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Kim Y, Ponomarenko J, Zhu Z, Tamang D, Wang P, Greenbaum J, Lundegaard C, Sette A, Lund O, Bourne PE. Immune epitope database analysis resource. Nucleic Acids Research. 2012;40(W1):W525–W530. doi: 10.1093/nar/gks438. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Kim Y, Sidney J, Pinilla C, Sette A, Peters B. Derivation of an amino acid similarity matrix for peptide: MHC binding and its application as a bayesian prior. BMC Bioinformatics. 2009;10(1):394. doi: 10.1186/1471-2105-10-394. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Larsen MV, Lelic A, Parsons R, Nielsen M, Hoof I, Lamberth K, Loeb MB, Buus S, Bramson J, Lund O. Identification of CD8 T cell epitopes in the west nile virus polyprotein by reverse-immunology using NetCTL. PloS One. 2010;5(9):e12697. doi: 10.1371/journal.pone.0012697. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lefranc MP, Giudicelli V, Ginestoux C, Bodmer J, Muller W, Bontrop R, Lemaitre M, Malik A, Barbie V, Chaume D. IMGT, the international ImMunoGeneTics database. Nucleic Acids Research. 1999;27(1):209–212. doi: 10.1093/nar/27.1.209. doi:gkc026. [pii] [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lin H, Zhang G, Tongchusak S, Reinherz EL, Brusic V. Evaluation of MHC-II peptide binding prediction servers: Applications for vaccine research BMC Bioinformatics. 2008;9(Suppl 12):S22. doi: 10.1186/1471-2105-9-S12-S22. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lund O, Nascimento EJ, Maciel Jr M, Nielsen M, Larsen MV, Lundegaard C, Harndahl M, Lamberth K, Buus S, Salmon J. Human leukocyte antigen (HLA) class I restricted epitope discovery in yellow fewer and dengue viruses: Importance of HLA binding strength. PloS One. 2011;6(10):e26494. doi: 10.1371/journal.pone.0026494. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lundegaard C, Hoof I, Lund O, Nielsen M. State of the art and challenges in sequence based T-cell epitope prediction. Immunome Research. 2010;6(Suppl 2):S3. doi: 10.1186/1745-7580-6-S2-S3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lundegaard C, Lamberth K, Harndahl M, Buus S, Lund O, Nielsen M. NetMHC-3.0: Accurate web accessible predictions of human, mouse and monkey MHC class I affinities for peptides of length 8–11. Nucleic Acids Research. 2008;36(suppl 2):W509–W512. doi: 10.1093/nar/gkn202. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lundegaard C, Lund O, Nielsen M. Accurate approximation method for prediction of class I MHC affinities for peptides of length 8, 10 and 11 using prediction tools trained on 9mers. Bioinformatics. 2008;24(11):1397–1398. doi: 10.1093/bioinformatics/btn128. [DOI] [PubMed] [Google Scholar]
- Lundegaard C, Nielsen M, Lund O. The validity of predicted T-cell epitopes. Trends in Biotechnology. 2006;24(12):537–538. doi: 10.1016/j.tibtech.2006.10.001. [DOI] [PubMed] [Google Scholar]
- Moise L, McMurry JA, Buus S, Frey S, Martin WD, De Groot AS. In silico-accelerated identification of conserved and immunogenic variola/vaccinia T-cell epitopes. Vaccine. 2009;27(46):6471–6479. doi: 10.1016/j.vaccine.2009.06.018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Moutaftsi M, Peters B, Pasquetto V, Tscharke DC, Sidney J, Bui H, Grey H, Sette A. A consensus epitope prediction approach identifies the breadth of murine TCD8 -cell responses to vaccinia virus. Nature Biotechnology. 2006;24(7):817–819. doi: 10.1038/nbt1215. [DOI] [PubMed] [Google Scholar]
- Murphy K, editor. Janeway's immunobiology. New York: Garland Science; 2011. [Google Scholar]
- Nielsen M, Lundegaard C, Worning P, Lauemøller SL, Lamberth K, Buus S, Brunak S, Lund O. Reliable prediction of T-cell epitopes using neural networks with novel sequence representations. Protein Science. 2003;12(5):1007–1017. doi: 10.1110/ps.0239403. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Nielsen M, Lund O. NN-align. an artificial neural network-based alignment algorithm for MHC class II peptide binding prediction. BMC Bioinformatics. 2009;10(1):296. doi: 10.1186/1471-2105-10-296. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Nielsen M, Lundegaard C, Blicher T, Lamberth K, Harndahl M, Justesen S, Roder G, Peters B, Sette A, Lund O. NetMHCpan, a method for quantitative predictions of peptide binding to any HLA-A and-B locus protein of known sequence. PloS One. 2007;2(8):e796. doi: 10.1371/journal.pone.0000796. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Nielsen M, Lundegaard C, Blicher T, Peters B, Sette A, Justesen S, Buus S, Lund O. Quantitative predictions of peptide binding to any HLA-DR molecule of known sequence: NetMHCIIpan. PLoS Computational Biology. 2008;4(7):e1000107. doi: 10.1371/journal.pcbi.1000107. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Nielsen M, Lundegaard C, Lund O. Prediction of MHC class II binding affinity using SMM-align, a novel stabilization matrix alignment method. BMC Bioinformatics. 2007;8(1):238. doi: 10.1186/1471-2105-8-238. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Oseroff C, Sidney J, Kotturi MF, Kolla R, Alam R, Broide DH, Wasserman SI, Weiskopf D, McKinney DM, Chung JL. Molecular determinants of T cell epitope recognition to the common timothy grass allergen. The Journal of Immunology. 2010;185(2):943–955. doi: 10.4049/jimmunol.1000405. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Parker KC, Bednarek MA, Coligan JE. Scheme for ranking potential HLA-A2 binding peptides based on independent binding of individual peptide side-chains. The Journal of Immunology. 1994;152(1):163–175. [PubMed] [Google Scholar]
- Paul S, Arlehamn CSL, Scriba TJ, Dillon MB, Oseroff C, Hinz D, McKinney DM, Pro SC, Sidney J, Peters B. Development and validation of a broad scheme for prediction of HLA class II restricted T cell epitopes. Journal of Immunological Methods. 2015 doi: 10.1016/j.jim.2015.03.022. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Paul S, Kolla RV, Sidney J, Weiskopf D, Fleri W, Kim Y, Peters B, Sette A. Evaluating the immunogenicity of protein drugs by applying in vitro MHC binding data and the immune epitope database and analysis resource. Clinical and Developmental Immunology. 20132013 doi: 10.1155/2013/467852. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Paul S, Weiskopf D, Angelo MA, Sidney J, Peters B, Sette A. HLA class I alleles are associated with peptide-binding repertoires of different size, affinity, and immunogenicity. Journal of Immunology (Baltimore, Md: 1950) 2013;191(12):5831–5839. doi: 10.4049/jimmunol.1302101. [doi] [DOI] [PMC free article] [PubMed] [Google Scholar]
- Peters B, Sidney J, Bourne P, Bui H, Buus S, Doh G, Fleri W, Kronenberg M, Kubo R, Lund O. The immune epitope database and analysis resource: From vision to blueprint. PLoS Biology. 2005;3(3) doi: 10.1371/journal.pbio.0030091. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rammensee HG, Bachmann J, Emmerich NPN, Bachor OA, Stevanovic S. SYFPEITHI: Database for MHC ligands and peptide motifs. Immunogenetics. 1999;50(3):213–219. doi: 10.1007/s002510050595. [DOI] [PubMed] [Google Scholar]
- Reche PA, Reinherz EL. Immunoinformatics. Springer; 2007. Prediction of peptide-MHC binding using profiles; pp. 185–200. [DOI] [PubMed] [Google Scholar]
- Robinson J, Halliwell JA, Hayhurst JD, Flicek P, Parham P, Marsh SG. The IPD and IMGT/HLA database: Allele variant databases. Nucleic Acids Research. 2015;43(Database issue):D423–31. doi: 10.1093/nar/gku1161. [doi] [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rudensky AY, Preston-Hurlburt P, Hong S, Barlow A, Janeway CA. Sequence analysis of peptides bound to MHC class II molecules. Nature a-z Index. 1991;353(6345):622–627. doi: 10.1038/353622a0. [DOI] [PubMed] [Google Scholar]
- Salimi N, Fleri W, Peters B, Sette A. Design and utilization of epitope-based databases and predictive tools. Immunogenetics. 2010;62(4):185–196. doi: 10.1007/s00251-010-0435-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sette A, Rappuoli R. Reverse vaccinology: Developing vaccines in the era of genomics. Immunity. 2010;33(4):530–541. doi: 10.1016/j.immuni.2010.09.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sette A, Vitiello A, Reherman B, Fowler P, Nayersina R, Kast WM, Melief C, Oseroff C, Yuan L, Ruppert J. The relationship between class I binding affinity and immunogenicity of potential cytotoxic T cell epitopes. The Journal of Immunology. 1994;153(12):5586–5592. [PubMed] [Google Scholar]
- Sidney J, Peters B, Frahm N, Brander C, Sette A. HLA class I supertypes: A revised and updated classification. BMC Immunol. 2008;9(1):1471–2172. doi: 10.1186/1471-2172-9-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sidney J, Assarsson E, Moore C, Ngo S, Pinilla C, Sette A, Peters B. Quantitative peptide binding motifs for 19 human and mouse MHC class I molecules derived using positional scanning combinatorial peptide libraries. Immunome Research. 2008;4(1):2. doi: 10.1186/1745-7580-4-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Singh H, Raghava G. ProPred: Prediction of HLA-DR binding sites. Bioinformatics. 2001;17(12):1236. doi: 10.1093/bioinformatics/17.12.1236. [DOI] [PubMed] [Google Scholar]
- Singh H, Raghava G. ProPred1: Prediction of promiscuous MHC class-I binding sites. Bioinformatics. 2003;19(8):1009–1014. doi: 10.1093/bioinformatics/btg108. [DOI] [PubMed] [Google Scholar]
- Stern LJ, Brown JH, Jardetzky TS, Gorga JC, Urban RG, Strominger JL, Wiley DC. Crystal structure of the human class II MHC protein HLA-DR1 complexed with an influenza virus peptide. 1994 doi: 10.1038/368215a0. [DOI] [PubMed] [Google Scholar]
- Sturniolo T, Bono E, Ding J, Raddrizzani L, Tuereci O, Sahin U, Braxenthaler M, Gallazzi F, Protti MP, Sinigaglia F. Generation of tissue-specific and promiscuous HLA ligand databases using DNA microarrays and virtual HLA class II matrices. Nature Biotechnology. 1999;17(6):555–561. doi: 10.1038/9858. [DOI] [PubMed] [Google Scholar]
- Sylvester-Hvid C, Nielsen M, Lamberth K, Røder G, Justesen S, Lundegaard C, Worning P, Thomadsen H, Lund O, Brunak S. SARS CTL vaccine Candidates—HLA supertype, Genome-Wide scanning and biochemical validation. Scandinavian Journal of Immunology. 2004;59(6):632–632. doi: 10.1111/j.0001-2815.2004.00221.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Tangri S, Mothé BR, Eisenbraun J, Sidney J, Southwood S, Briggs K, Zinckgraf J, Bilsel P, Newman M, Chesnut R. Rationally engineered therapeutic proteins with reduced immunogenicity. The Journal of Immunology. 2005;174(6):3187–3196. doi: 10.4049/jimmunol.174.6.3187. [DOI] [PubMed] [Google Scholar]
- Trolle T, Metushi IG, Greenbaum JA, Kim Y, Sidney J, Lund O, Sette A, Peters B, Nielsen M. Automated benchmarking of peptide-MHC class I binding predictions. Bioinformatics (Oxford, England) 2015;31(13):2174–2181. doi: 10.1093/bioinformatics/btv123. [doi] [DOI] [PMC free article] [PubMed] [Google Scholar]
- Vincenti D, Carrara S, De Mori P, Pucillo LP, Petrosillo N, Palmieri F, Armignacco O, Ippolito G, Girardi E, Amicosante M, Goletti D. Identification of early secretory antigen target-6 epitopes for the immunodiagnosis of active tuberculosis. Molecular Medicine (Cambridge, Mass) 2003;9(3-4):105–111. [PMC free article] [PubMed] [Google Scholar]
- Wang M, Lamberth K, Harndahl M, Røder G, Stryhn A, Larsen MV, Nielsen M, Lundegaard C, Tang ST, Dziegiel MH. CTL epitopes for influenza A including the H5N1 bird flu; genome-, pathogen-, and HLA-wide screening. Vaccine. 2007;25(15):2823–2831. doi: 10.1016/j.vaccine.2006.12.038. [DOI] [PubMed] [Google Scholar]
- Wang P, Sidney J, Dow C, Mothé B, Sette A, Peters B. A systematic assessment of MHC class II peptide binding predictions and evaluation of a consensus approach. PLoS Computational Biology. 2008;4(4):e1000048. doi: 10.1371/journal.pcbi.1000048. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wang P, Sidney J, Kim Y, Sette A, Lund O, Nielsen M, Peters B. Peptide binding predictions for HLA DR, DP and DQ molecules. BMC Bioinformatics. 2010;11(1):568. doi: 10.1186/1471-2105-11-568. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Weiskopf D, Angelo MA, de Azeredo EL, Sidney J, Greenbaum JA, Fernando AN, Broadwater A, Kolla RV, De Silva AD, de Silva AM, Mattia KA, Doranz BJ, Grey HM, Shresta S, Peters B, Sette A. Comprehensive analysis of dengue virus-specific responses supports an HLA-linked protective role for CD8+ T cells. Proceedings of the National Academy of Sciences of the United States of America. 2013;110(22):E2046–53. doi: 10.1073/pnas.1305227110. [doi] [DOI] [PMC free article] [PubMed] [Google Scholar]
- Zhang GL, Khan AM, Srinivasan KN, August JT, Brusic V. MULTIPRED: A computational system for prediction of promiscuous HLA binding peptides. Nucleic Acids Research. 2005;33(suppl 2):W172–W179. doi: 10.1093/nar/gki452. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Zhang L, Udaka K, Mamitsuka H, Zhu S. Toward more accurate pan-specific MHC-peptide binding prediction: A review of current methods and tools. Briefings in Bioinformatics. 2012;13(3):350–364. doi: 10.1093/bib/bbr060. [doi] [DOI] [PubMed] [Google Scholar]
- Zhang Q, Wang P, Kim Y, Haste-Andersen P, Beaver J, Bourne PE, Bui HH, Buus S, Frankild S, Greenbaum J, Lund O, Lundegaard C, Nielsen M, Ponomarenko J, Sette A, Zhu Z, Peters B. Immune epitope database analysis resource (IEDB-AR) Nucleic Acids Research. 2008;36(Web Server issue):W513–8. doi: 10.1093/nar/gkn254. [doi] [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.