Statistical Analysis of Metagenomic Data *SLIDES NOW AVAILABLE*

Ian Clark, Michael Defoin-Platel, Elisa Loza, Wally Gilks (Rothamsted Research)

Suparna Mitra (Tuebingen University)

Metagenomics is a rapidly developing science, promising expansion towards discoveries that can help in the comprehension, cure and prevention of many diseases, in monitoring the impact of pollutants on ecosystems and in mining the rich genetic resource of non-culturable microbes that may lead to the discovery of new genes, enzymes, and natural products.

The recent development of new, less expensive, ultra-high throughput sequencing technologies that can produce huge numbers of DNA reads at an affordable cost, has boosted the number and scope of metagenomic sequencing projects. It has resulted into a dramatic increase in the volume of sequence data that must be analyzed. The analysis of metagenomic datasets is an immense conceptual and computational challenge.

The analysis  often starts by asking the questions of "who is out there?', "what are they doing?" and "how do they compare?". This talk will briefly describe, how these computational questions can be addressed using MEGAN, the MEtaGenome ANalyzer program. First how to analyze the taxonomic and functional content of a single dataset and then more specially showing how such analyses can be performed in a comparative fashion. I will demonstrate how to compare different datasets using ecological indices and other distance measures. The discussion will be conducted using a number of published marine datasets comprising metagenomic, metatranscriptomic and 16S rRNA data.
 

Vincent Plagnol (University College London)

It is becoming increasingly clear that human pathogens play a role in multiple disorders which do not have an obvious infectious basis. This is in particular the case for autoimmune disorders such as type 1 diabetes as the interplay between the host immune system and infectious agents shapes our immunity and has a long term effect on human biology. All active pathogens, and in particular viruses, leave a RNA signature in affected tissues. The advent of high throughput DNA sequencing techniques, in particular transcriptome sequencing or RNA-Seq, is an opportunity to interrogate human tissues for the presence of these infectious agents. However, the use of short sequencing reads limits the sensitivity and specificity of viral identification. To overcome these issues, it is necessary to efficiently combine read assembly with homology/blast based methods in order to increase read length and provide a more powerful tool for viral sequence detection. In this talk, I will present an ongoing methodological work to address this question. Using a combination of simulations and actual RNA-Seq data, I will highlight the reasons why metagenomic analysis of short read sequence data is challenging and show how these limitations can be overcome.