摘要：Metagenomics binning process is a step prior to the taxonomic assignment of metagenomic reads of contigs, which helps to group genome sequences belonging to the same species. In this paper we propose a clustering method that is executed recursively to cluster contigs into groups of same taxa. In each step the method increases the taxonomic level, beginning with a domain and ending with a group that represents the species. The method uses a previous rule-based system to separate virus from the rest of the organism and feature selection algorithms to select different features in each step of the clustering. The clustering is based on k-means++ using Cosine and Jaccard distance, and feature selection on gain information. The proposed method outperforms classic k-means++, achieving 88.15% of purity in clusters.
