European Journal of Biomedical and Pharmaceutical Sciences

AGPRED: ANTIGEN PREDICTION USING NEURAL NETWORK BASED ON TRI-PEPTIDE MARKERS

Himanshu S. Mazumdar* and Ragini V. Oza

ABSTRACT

The protein sequence plays an important role to understand the function and feature of protein. Antigen prediction from the huge amount of protein primary sequence is a challenging problem. A novel approach is proposed here to characterize an antigen sequence using a set of features which can describe characteristics of target antigen group. The proposed system uses a combination of evolutionary algorithm and proposed ordering algorithm to identify the set features. Here the features are the tri-peptides. An algorithm ensures that the unique combination of the tri-peptide separates target antigen sequence from other protein sequences. We have preprocessed the datasets of Plasmodium Falciparum, Leptospira Interrogans, Pseudomonas Aeruginosa, Streptococcus Pneumonia and Bacillus Thuringiensis to use it in our system. These datasets are extracted from Uniref100 protein sequence database which contains 83 million records. Neural networks are trained using training sets from all species and results are compared here. A prediction result gives 98 % of peak accuracy for P.Falciparum using identified tri-peptide features which are tested on the test dataset.

Keywords: Plasmodium Falciparum; Tri-peptide Residue; Occurrence Frequency; Population Ratio; Genetic Algorithm; Back-propagation Neural Network.