KISS: Kernel-based Inter-allele peptide binding prediction SyStem


KISS predicts whether or not a 9-mer will bind an MHC-I molecule for various alleles.


The first number next to each allele name is the number of epitopes that were available for the allele during the training.
The second number is the mean success rate observed on the 5-folded data that was used to build the classifier. At 1, the classifier made no mistake, at 0.5 the classifier made random-like predictions.


Enter your data:

Use Ctrl+left-click for multiple selection.

How does it work?

The binding prediction is made by an SVM using a multitask kernel to leverage the available training information across the alleles, which improves its accuracy, especially for the alleles with few known epitopes.

The predictor is trained on this database which contains known epitopes from syfpeithi, mhcbn, lanl and iedb databases.

More details on the model and its performance are available in the reference below.

Reference

Efficient peptide-MHC-I binding prediction for alleles with few known binders, L. Jacob and J.-P. Vert, Bioinformatics, 2008 (to appear)

Available here.

Contact: Laurent Jacob.

Acknowledgement

This server was developed by the Center for Computational Biology of Ecole des Mines de Paris.


ENSMP logo

© Laurent Jacob

Valid XHTML 1.1 and CSS