Protein-protein Interaction Prediction Based On Computational Intelligence

Each year the number of protein sequences is rapidly increasing while known to slowgrowth in the number of protein structure, therefore an urgent need to develop a rapid andaccurate calculation tool to predict protein interactions. This paper revolves around severalimportant aspects of protein interaction prediction: protein feature extraction methods,machine learning algorithms and integrated learning algorithm study, purpose is to get a quickand effective method of prediction of protein interactions.Using machine learning algorithm to predict protein interactions, is essentially a patternrecognition problem. We study a basic assumption is that protein interaction is determined byits amino acid sequence only, and for the same kind of protein, the amino acid sequence hassome inherent regularity, the inherent law in mathematical formula is very difficult to express.Using machine learning methods for protein interaction prediction, and is a supervisedlearning process, by category of protein sequences known samples to train neural network,support vector machine (SVM) and Bayesian neural network (such as machine learning model,make its study in internal regularity of protein sequences, allowing it to meet unknowncategories of proteins can make scientific and reasonable judgment.And other pattern recognition problems, the amino acid sequence feature extraction is touse machine learning algorithm is a priority for the protein interaction prediction. Featureextraction is to use letters into the sequence of amino acids has a fixed dimension data vector,so that the computer for processing. Amino acids feature extraction is a very important part ofthe protein interaction prediction, feature extraction method is proper for model predictionaccuracy has crucial effect. Protein feature extraction method has a lot of, mainly, dipeptidemodel, polypeptide, amino acid composition model, pseudo amino acid composition (PseAA),physical and chemical properties model (PCC) and quantification analysis (RQA), etc. In thispaper, through the best-first features screening strategy to get a new combination: model andrecreate the quantitative analysis of the physical and chemical composition, and on thecharacteristics of fusion.Methods two classifiers are commonly used K nearest neighbor method (KNN), the Biasnetwork, artificial neural network (ANN), the flexible neural tree (FNT) etc.. But the prediction of protein-protein interaction is a typical classification problems, several classifierswith the above mentioned on the simple classification, the prediction results are not ideal.Therefore, the classification problem in multi feature under the time. In this paper, the Humandata set of protein protein interaction prediction, classification using artificial neural network,optimization algorithm of network coefficient by using the particle swarm optimizationalgorithm (PSO), a better prediction result achieved by this combination.This paper using the auto covariance encoding and resonant recognition modelconstructing classification model, which using the auto covariance coding classificationmodel in the Human has achieved83.5%accuracy on data sets, resonant recognition model inthe Human data set were achieved81.9%accuracy. It also proved the effectiveness of thework.