New Approaches to Improving Organisms Detection and Gene Prediction in Metagenomes.

摘要

This thesis investigates microorganism detection and gene prediction on the DNA level from environmental samples such as air, water, soil, and the human body. In applications such as homeland security, it is important to detect pathogens and their expressed genes in the air or ground. Chemical solutions have difficulty detecting trace amounts and are costly. Technologies such as DNA microarrays and next-generation sequencing are becoming cost-efficient, thus it is of interest to use these technologies to discover organisms and genes present in a sample, especially when in “fine” amounts. The thesis will impact organism and gene discovery, which will aid pharmaceutical and bio-fuel discovery, as well as new methods in forensics and homeland security.;To enable microarrays to detect more organisms utilizing fewer probes, we discuss the advantage of using specially-designed compressive sensing microarrays (GSM), which is a DNA-based sensor array that operates using group testing and compressive sensing (CS) principles. We improve compressive sensing DNA microarrays by introducing probe picking algorithm that iteratively searches for the most optimal probe.;For predicting reads from next-generation sequencing of DNA, we introduce a new hybrid algorithm, Homology-Abinitio, which combines ab-initio annotation, using improved models, with sequence similarity comparisons. After benchmarking well-known gene prediction programs for metagenomic reads—GeneMark, MetaGeneAnnotator (MGA) and Orphelia—we then combined them to gain better accuracy (f-measure), especially for short reads. In addition, we show how using metatranscriptomic data can significantly improve gene prediction specificity. Each previous algorithm and Homology-Abinitio is tested with the most diverse test set to-date, including genes from 96 species.