NetCTL - 1.2

Predection of CTL epitopes in protein sequences

The NetCTL 1.2 server predicts CTL epitopes in protein sequences. The current version 1.2 is an update to the version 1.0. Version 1.2 expands the MHC class I binding predicition to 12 MHC supertypes including the supertypes A26 and B39. The accuracy of the MHC class I peptide binding affinity is significantly improved compared to the earlier version. Also the prediction of proteasonal cleavage has been improved and is now identical to the predictions obtained by the NetChop-3.0 server. The updated version has been trained on a set of 886 known MHC class I ligands.

NOTE. On Aug 16 2006 a minor update to the server has been implemented improving the prediction accuracy for MHC binding. The earlier version of the NetCTL 1.2 server (1.2 beta) is available via the versions history for the server.

The method integrates prediction of peptide MHC class I binding, proteasomal C terminal cleavage and TAP transport efficiency. The server allows for predictions of CTL epitopes restricted to 12 MHC class I supertype. MHC class I binding and proteasomal cleavage is performed using artificial neural networks. TAP transport efficiency is predictied using weight matrix.

The MHC peptide binding is predicted using neural networks trained as described for the NetMHC server. The proteasome cleavage event is predicted using the version of the NetChop neural networks trained on C terminals of known CTL epitopes as describe for the NetChop-3.0 server. The TAP transport efficiency is predicted using the weight matrix based method describe by Peters et al.

The server includes predictions of MHC/peptide binding for 12 MHC class I supertypes. The output from the neural network predicting MHC/peptide binding is a log transformed value related to the IC50 values in nM units. For details on the transformation please see the Output format

The scores from the three individual prediction methods are integrated as a weighted sum with a relative weight on peptide/MHC binding of 1. Different thresholds for the integrated score can be translated into sensitivity/specificity values. In a large benchmark calculate containing more than 800 known MHC class I ligands the following relations were found

Score Sensitivity Specificity
> 1.25 0.54 0.993
> 1.00 0.70 0.985
> 0.90 0.74 0.980
> 0.75 0.80 0.970
> 0.50 0.89 0.940

The project is collaboration between CBS and IMMI at Copenhagen university


Paste a single sequence or several sequences in FASTA format into the field below:

Submit a file in FASTA format directly from your local disk:


Weight on C terminal cleavage                
Weight on TAP transport efficiency        
Threshold for epitope identification        

Sort by score  

At most 5000 sequences per submission; each sequence not more than 20,000 amino acids and not less than 9 amino acids for at most 200,000 amino acids in total.

The sequences are kept confidential and will be deleted after processing.


For publication of results, please cite:

  • Current version:

    Large-scale validation of methods for cytotoxic T-lymphocyte epitope prediction.
    Larsen MV, Lundegaard C, Lamberth K, Buus S, Lund O, Nielsen M.
    BMC Bioinformatics. Oct 31;8:424. 2007

    View the abstract        

  • Earlier versions:

    An integrative approach to CTL epitope prediction. A combined algorithm integrating MHC-I binding, TAP transport efficiency, and proteasomal cleavage predictions.
    Larsen M.V., Lundegaard C., Kasper Lamberth, Buus S,. Brunak S., Lund O., and Nielsen M.
    European Journal of Immunology. 35(8): 2295-303. 2005

    View the abstract        

  • Related publications:

    Reliable prediction of T-cell epitopes using neural networks with novel sequence representations.
    Nielsen M, Lundegaard C, Worning P, Lauemoller SL, Lamberth K, Buus S, Brunak S, Lund O.
    Protein Sci., 12:1007-17, 2003.

    View the abstract

    Sensitive quantitative predictions of peptide-MHC binding by a 'Query by Committee' artificial neural network approach.
    Buus S, Lauemoller SL, Worning P, Kesmir C, Frimurer T, Corbet S, Fomsgaard A, Hilden J, Holm A, Brunak S.
    Tissue Antigens., 62(5):378-84, 2003.

    View the abstract

    The role of the proteasome in generating cytotoxic T cell epitopes: Insights obtained from improved predictions of proteasomal cleavage.
    M. Nielsen, C. Lundegaard, S. Brunak, O. Lund, and C. Kesmir. Immunogenetics., 57(1-2):33-41, 2005.

    View the abstract

    Identifying MHC class I epitopes by predicting the TAP transport efficiency of epitope precursors
    Peters, B., Bulik, S., Tampe, R., Endert, P. M. V. and Holzhutter, H. G.
    J. Immunol. 171: 1741-1749, 2003.

    View the abstract


In order to use the NetCTL server for prediction on amino acid sequences:
  1. Enter the sequence in the sequence window, or give a file name.

    The sequence must be written using the one letter amino acid code: `acdefghiklmnpqrstvwy' or `ACDEFGHIKLMNPQRSTVWY'.
    Other letters will be converted to `X' and treated as unknown amino acids.
    Other characters, such as whitespace and numbers, will simply be ignored.

  2. Select HLA supertype (10 HLA supertypes are available)

  3. Optional
    1. Weight on C terminal cleavage. The default select value (0.1) gives optimal predictive performance on average. You can modify the relative weight on proteasomal cleavage by entering a different weight value.
    2. Weight on TAP transport efficiency. The default select value (0.05) gives optimal predictive performance on average. You can modify the relative weight on TAP transport by entering a different weight value.
    3. Threshold for epitope identification. Peptides with a combined prediction score value greater than the threshold value are marked as potential epitopes. In a large scale benchmark identifying known CTL epitope in proteins the default value of 0.75 was found to correspond to a sensitivity of 0.65 and a specificity 0.97. Note that the benchmark is highly unbalanced since only one peptide is identified as CTL epitope in each protein, and the number of negatives hence is orders of magnitude larger than the number of positives. This has important implications for the interpretation of the specificity values.
    4. Sort by score. Select which feature to sort on. Sorting is done in decreasing order. Default is no sorting.

  4. Press the "Submit sequence" button.

  5. A WWW page will return the results when the prediction is ready. Response time depends on system load.

Output format


The output consists of 16 columns:
  • Residue number.
  • ID
  • Protien identifier
  • pep
  • Peptide sequence
  • aff
  • Predicted MHC binding affinity. The value is give as 1 - log50k(aff), where log50k is the logaritm with base 50.000, and aff is the affinity in nM units
  • aff_rescale
  • Rescale binding affinity. The predicted binding affinity is normalized by the 1st percentile score
  • cle
  • C terminal cleavage affinity
  • tap
  • TAP transport efficiency
  • COMB
  • Prediction score
  • <-E. Identified MHC ligands


    NetCTL predictions using MHC supertype A24. Threshold 0.500000
      17 ID 143B_BOVIN pep AERYDDMAA aff   0.0349 aff_rescale   0.0478 cle 0.0344 tap  -0.5270 COMB   0.0249 
      18 ID 143B_BOVIN pep ERYDDMAAA aff   0.0348 aff_rescale   0.0476 cle 0.4333 tap  -0.1920 COMB   0.0813 
      19 ID 143B_BOVIN pep RYDDMAAAM aff   0.3017 aff_rescale   0.4128 cle 0.9943 tap   0.4700 COMB   0.5357 <-E
      20 ID 143B_BOVIN pep YDDMAAAMK aff   0.0352 aff_rescale   0.0482 cle 0.9622 tap   0.0930 COMB   0.1491 
      21 ID 143B_BOVIN pep DDMAAAMKA aff   0.0343 aff_rescale   0.0469 cle 0.6469 tap  -1.1460 COMB   0.0543
    Number of MHC ligands 8 identified. Number of amino acids 237. Protein name 143B_BOVIN

  • Supplementary material for update to the NetCTL method (v. 1.2):

    Original method

    An integrative approach to CTL epitope prediction. A combined algorithm integrating MHC-I binding, TAP transport efficiency, and proteasomal cleavage predictions.

    The two data sets used in the benchmark calculation are given below in the FASTA format. For each entry is given the Swiss-Prot name of the protein "hosting" the epitope, the epitope position, the epitope sequence, and the HLA supertype.
    SYFPEITHI dataset
    HIV dataset
    HIV_EpiJen dataset
    List of known CTL epitopes from SYFPEITHI and the Los Alamos HIV databases
    Difference between the A1, A2, and A3 epitope-protein pairs for the HIV and HIV_EpiJen datasets
    Supplementary figure: Comparing specificities on the HIV EpiJen dataset


    M. V. Larsen, C. Lundegaard, K. Lamberth, S. Buus, S. Brunak, O. Lund, and M. Nielsen An integrative approach to CTL epitope prediction. A combined algorithm integrating MHC binding, TAP transport efficiency, and proteasome cleavage predictions. European Journal of Immunology 2005 Aug;35(8):2295-303.



    Large-scale validation of methods for cytotoxic T-lymphocyte epitope prediction. Larsen MV, Lundegaard C, Lamberth K, Buus S, Lund O, Nielsen M. BMC Bioinformatics. 2007 Oct 31;8:424.


    BACKGROUND: Reliable predictions of Cytotoxic T lymphocyte (CTL) epitopes are essential for rational vaccine design. Most importantly, they can minimize the experimental effort needed to identify epitopes. NetCTL is a web-based tool designed for predicting human CTL epitopes in any given protein. It does so by integrating predictions of proteasomal cleavage, TAP transport efficiency, and MHC class I affinity. At least four other methods have been developed recently that likewise attempt to predict CTL epitopes: EpiJen, MAPPP, MHC-pathway, and WAPP. In order to compare the performance of prediction methods, objective benchmarks and standardized performance measures are needed. Here, we develop such large-scale benchmark and corresponding performance measures and report the performance of an updated version 1.2 of NetCTL in comparison with the four other methods.

    RESULTS: We define a number of performance measures that can handle the different types of output data from the five methods. We use two evaluation datasets consisting of known HIV CTL epitopes and their source proteins. The source proteins are split into all possible 9 mers and except for annotated epitopes; all other 9 mers are considered non-epitopes. In the RANK measure, we compare two methods at a time and count how often each of the methods rank the epitope highest. In another measure, we find the specificity of the methods at three predefined sensitivity values. Lastly, for each method, we calculate the percentage of known epitopes that rank within the 5% peptides with the highest predicted score.

    CONCLUSION: NetCTL-1.2 is demonstrated to have a higher predictive performance than EpiJen, MAPPP, MHC-pathway, and WAPP on all performance measures. The higher performance of NetCTL-1.2 as compared to EpiJen and MHC-pathway is, however, not statistically significant on all measures. In the large-scale benchmark calculation consisting of 216 known HIV epitopes covering all 12 recognized HLA supertypes, the NetCTL-1.2 method was shown to have a sensitivity among the 5% top-scoring peptides above 0.72. On this dataset, the best of the other methods achieved a sensitivity of 0.64. The NetCTL-1.2 method is available at All used datasets are available at

    Software Downloads


    Correspondence:        Technical Support: