Targeted Computational Approaches for Mining Functional Elements in Metagenomes

Abstract

Metagenomics enables the genomic study of uncultured microorganisms by directly extracting the genetic material from microbial communities for sequencing. Fueled by the rapid development of Next Generation Sequencing (NGS) technology, metagenomics research has been revolutionizing the field of microbiology, revealing the taxonomic and functional composition of many microbial communities and their impacts on almost every aspect of life on Earth. Analyzing metagenomes (a metagenome is the collection of genomic sequences of an entire microbial community) is challenging: metagenomic sequences are often extremely short and therefore lack genomic contexts needed for annotating functional elements, while whole-metagenome assemblies are often poor because a metagenomic dataset contains reads from many different species. Novel computational approaches are still needed to get the most out of the metagenomes.

In this dissertation, I first developed a binning algorithm (AbundanceBin) for clustering metagenomic sequences into groups, each containing sequences from species of similar abundances. AbundanceBin provides accurate estimations of the abundances of the species in a microbial community and their genome sizes. Application of AbundanceBin prior to assembly results in better assemblies of metagenomes--an outcome crucial to downstream analyses of metagenomic datasets.

In addition, I designed three targeted computational approaches for assembling and annotating protein coding genes and other functional elements from metagenomic sequences. GeneStitch is an approach for gene assembly by connecting gene fragments scattered in different contigs into longer genes with the guidance of reference genes. I also developed two specialized assembly methods: the targeted-assembly method for assembling CRISPRs (Clustered Regularly Interspersed Short Palindromic Repeats), and the constrained-assembly method for retrieving chromosomal integrons. Applications of these methods to the Human Microbiome Project (HMP) datasets show that human microbiomes are extremely dynamic, reflecting the interactions between community members (including bacteria and viruses).