Researches On MHC ? Binding Peptides Prediction

The major histocompatibility complex II(MHC II)molecules play a central role in human immune system.By presenting the binding peptide to helper T cells,MHC II molecules trigger adaptive immune response to infections.Therefore,accurate prediction of MHC II binding peptides can help to understand the mechanism of immune systems and promote the development of precision medicine.Feature extraction is the major challenge for the binding peptide prediction.This paper focused on the binding peptide prediction and proposed a new MHC II binding peptide prediction method called MHC2 NNZ based on integrated artificial neural networks.MHC2 NNZ contained two new feature extraction methods.The experiments on benchmark dataset showed that our algorithm had good performance.The main innovations of this thesis was included:Firstly,we proposed a new method of peptide flanking regions(PFR)feature extraction.The new PFR feature extraction method applied z descriptors to extract features of PFR,thus a novel representation way of PFR was formed.Then,MHC2 NNZ constructed the prediction model based on integrated artificial neural networks to recognize unknown binding peptides.In this paper,the benchmark dataset IEDB SR was used to evaluate MHC2 NNZ based on the new PFR feature extraction method.This MHC2 NNZ achieved good performance on human MHC II molecules DQ(HLA DQ),which was generally better than the classical algorithms.The results of MHC2 NNZ based on the new PFR feature extraction method demonstrated that the PFR contribute less to DQ molecules peptide binding compared to DR molecules and DP molecules.Secondly,we proposed a new method of anchor sites feature extraction.The new anchor sites feature extraction method used the physicochemical features of HQI to extract features of anchor sites of the peptide binding core,thus became the feature extraction way mainly based on physicochemical properties.Then,MHC2 NNZ built integrated artificial neural network model and this model was evaluated on benchmark dataset IEDB SR.MHC2 NNZ based on the new anchor sites feature extraction method outperformed than all classical algorithms on HLA DR molecules and HLA DQ molecules.The experimental results demonstrated that physicochemical properties can play more important roles in solving machine learning problems such as overfitting in MHC II binding peptides prediction.