Prediction Of Cytokine Families Based On Multiple Sequence Features

Recognition and classification of protein family are one of the most important missions in post-genomic era. Cytokines are a kind of protein that are produced by immunocytes or related cells to regulate functions of certain cells. They play important roles in many physiological activities of human. The prediction of cytokine families, especially those cytokines whose functions are unknown, not only helps to reveal the pathological or physiological transformation mechanism of a body, but also makes a direct guide to such application fields as biological pharmacy and disease treatment. Therefore it is of great importance to do this research. However, in face of the increasing data of protein sequences, it is always difficult to find an effective computational method to predict protein families and determine their functions, which is still one of the most challenges for bioinformatics research.This paper first analyzes the existing prediction method of protein, and illustrates multiple feature extraction methods including amino acid & dipeptide composition, dipeptide composition & length, and pseudo amino acid, using the support vector machine tool in machine learning theory. Then, a prediction server called CytoKey for cytokine family classification and recognition are developed in this study. According to the results of comparison test, CytoKey shows a significant improvement in the accuracy of cytokine prediction, in contrast to former methods. And the dipeptide composition & length feature method is much better than the latest released cytokine prediction software CTKPred. CytoKey is available through the Internet at http://med-computing.com.