Network Inference of microbial communities from metagenome data

• 作 者: Huaying Fang Chengcheng Huang Hongyu Zhao Minghua Deng
• 机 构: LMAM,SchoolofMathematicalSciences,PekingUniversity,Beijing100871,China;BeijingInternationalCenterfor
• 出 处: 第七届全国生物信息学与系统生物学学术大会
• 发表日期: 2016年10期

　　Microbes play an important role in the environment and human life.Bacteria and archaea have been found in extreme conditions such as deep sea vents with high temperatures and rocks of boreholes beneath the Earths surface.Microbes affect the human life on our food,health and medicine.The way in which microbes affect the human health remains largely unknown.Analysis of the human microbiome may help us better understand our own.genome.The high throughput sequencing technologies,such as 16S rRNA gene profiling,provide an uncultivated microbial sampling strategy for diverse natural microbe communities.However,sequencing data only represent the relative abundances of microbial species owing to different collection scales and various sequencing depths.This feature of microbiome data is called compositionality.Statistical analysis of such compositional data presents unique challenges since the constant sums restriction can bring spurious results if it is ignored.Another challenge is that the sample size is generally less than the number of species.We propose a series of statistical methods,including CCLasso,gCoda and their extension,to deal with sparse network inference problem from metagenome data.Our methods directly estimate the covariance and precision matrix among among the latent variables for the compositional data.Our methods work much better than existing methods,such as SparCC and SPIEC-EASI in both simulation and real data.