Using Direct and Indirect Neighbours to Predict Protein Function in GO-Evaluated PPI Data Set

The recent development of high-throughout techniques to generate large volumes of protein-protein interaction(PPI) data, which increased the need for methods that annotate the function of protein. Some methods use indirect method to predict proteins function. However, due to the nature of noise, the relationship between proteins may not be existed in truth. In this paper, we propose a method of protein function prediction in GO-evaluated PPI data set. Firstly, the original PPI data set is evaluated by protein similarity method based on GO. Secondly, we develop an algorithm, FAW, which takes into account both direct and indirect functional association, to predict the function of proteins. Our approach is evaluated on four human PPI data sets. The experimental results show our approach has good performance in terms of efficiency.