hej hej


Prediction

Information

Back to SBS

SVMHC development

The first version of SVMHC was developed at the Stockholm Bioinformatics Center, Sweden. SVMHC is currently maintained at the Division for Simulation of Biological Systems, University of Tubingen, Germany. The current version of SVMHC enables prediction of both MHC class I and MHC class II binding peptides. The graphical output also allows for simple identification of promiscuous epitopes.
For further information about SVMHC, please contact Pierre Dönnes.

How to cite the prediction service

For publications citing the SVMHC method, please use:
Prediction of MHC class I binding peptides, using SVMHC. Pierre Dönnes and Arne Elofsson in: BMC Bioinformatics 2002 3: 25

Input format

There are two ways to submit data for prediction. Sequences can be pasted in Fasta format or specified by
database ids. In the case of database ids, both SWISSPROT accessions (e.g. P18146 or EGR1_HUMAN) and NCBI
RefSeq accessions (e.g. NP_055147) can be used. Please specify a fasta sequence OR a database id.

MHC class I prediction

The MHC class I predictions are based on support vector machines (SVMs) and known MHC-binding peptides. The sequences of peptides binding to different MHC alleles were extracted from the MHCPEP and SYFPEITHI databases. The SYFPEITHI databases only contain peptides that are naturally processed and presented, whereas the MHCPEP also contain synthetic peptides. For further details about data extraction and SVM training, see the original SVMHC paper.

MHC class II prediction

The current version of SVMHC also enables prediction of MHC class II binding peptides. The matrices used were published Sturniolo et al. and are used by the TEPITOPE software. Please note that these values were taken from the original paper without any updates of the matrices considered.

Output formats format

  • The new version of SVMHC offers multiple ways to analyze the prediction results. The first view is a graphical overview of the predicted MHC binding peptides for each allele. This view can be used to search for peptides shared over different alleles, something a bit more hard to find from simple rank tables. The first output page also displays ranked lists of predicted binders(with score >0). The scores are typically in the range from -1 to +1, where higher scores can be interpreted as more likely binders. Unpublished stidues have also verified that SVMHC performs reasonably well for quantitative prediction.
  • An additional view is the SUMMARY TABLES. These tables are split into octamer, nonamer, and decamer peptides. The tables gives all possible mers of the protein along with the predicted scores for each allele.
  • The ALL RANKED option gives the complete lists of ranked peptides for each allele.
  • The SUMMARY TABLES TAB is gives a tab separated file of the summary tables back. This file can be opened in e.g. Microsoft Excel for furhter analysis.
  • If a databse id was used for prediction, links to the relevant database entry is also given.