Abstract:Metagenomic binning is a fundamental question for metagenomic studies. Features extraction is the main factor which influences the performance of metagenomic binning, and how to extract the appropriate feature vectors will influence the binning accuracy and running time. Therefore, this paper proposes a features extraction method which based on third-order Markov model and transferring probability matrix for metagenomic binning problem. Meanwhile, we employ the features selection method based on mutual information to reduce the dimensions of feature vectors and apply it to support vector machine algorithm for binning as well as making comparisons among similar binning algorithms. The results show that this new features extraction method possesses applicable discriminability among different metagenomic species, which is particularly appropriate for large-scale metagenomic binning problem.