A server for predicting and designing
interferon-gamma inducing epitopes

Algorithm of IFNepitope

In past various methods have been develope which predict T-helper epitopes directly or indirectly (MHC Class II binder). Best of our knowledge no method has been developed for predicting epitope or antigenic peptides that can induce IFN-gamma. First time an attempt have been made to develop method for predicting IFN epitopes.

First we collect all MHC class II binders from IEDB database. We remove all those binders whose IFN-gamma inducing potential has not been tested. Finally we got 3705 IFN-gamma inducing and 6728 non-inducing MHC class II binders (See flowchart). This dataset is used for developing models for predicting IFN-gamma epitopes.

This algorithm of the server is relies on the following three models:

Motif based model

Aim of this approach is to discover motifs or patterns in IFN-gamma inducing and non-inducing MHC class II binders. In this study we used a powerful pattern discovery software MERCI. First we identified motifs are exclusively found in IFN-gamma inducing peptdes and not present in non-inducing peptides. Similarly we identified motifs in non-inducing peptides which are absent in IFN-gamma inducing peptides. These novel motifs used for discriminating IFN-gamma inducing and non-inducing peptides.

SVM based models

In this study, Support Vector Machine (SVM)based models has been developed using software SVM_Light. Major fearures used for training and testing SVM models residue composition of peptides that include amino acid and dipeptide composition.

Hybrid approach(Motif and SVM)

In hybrid model, we combine strength of both techniques motif based and SVM based for discriminating IFN-gamma inducing peptides. For a query peptide first we check whether it contain any known peptides disovered in our study, if yes we assign type of peptide based on type of motif it contain. If no motif is found then we use SVM model for prediction.