A novel gene identification method in metagenomic shotgun reads

摘要

@@The newly-developed metagenomic approach which obtains shotgun DNA sequences directly from environments is rapidly becoming a powerful approach for studying the microorganisms in natural niches and microbiomes resided in and on human bodies. As many sequences remain as unassembled one-pass reads, most conventional gene-finding tools are failure to identify genes in metagenomic samples. Although similarity-based method is widely used currently, it's also questioned for its disability in finding novel genes. Up to now, several algorithms were desired for ab initio metagenomic gene identification. MetaGene and MetaGeneMark construct regressive models between frequencies of oligonucleotides in protein-coding regions and genomic nucleotide composition, while Orphelia uses a large scale machine-learning method with a combination of neural networks and linear discriminates. Here, we present a novel method, MetaGun, which comprises a supervised universal model and a data-specific novel model based on SVM architectre.