Seminars (MTG)

In my talk I will outline the main reasons to be pessimistic about metagenomic assembly and discuss possible ways forward. In particular, I will argue that 'metagenomic assembly' is an ill defined concept and that we, as a community, need to identify and formalize specific use cases that address relevant biological problems and that are computationally tractable. I will also discuss issues related to the validation of metagenomic assemblies, an important yet often overlooked problem.

Most standard tools for assembly of next generation sequencing reads are not designed for metagenomic datasets, rather the aim is to assemble a single genome into a relatively small number of large contigs or scaffolds. In part this is achieved by implementation of algorithms designed to remove sequencing errors and to smooth small variations that would otherwise break contiguity. However, this approach has limits when used to analyse environmental samples, in which fragments of DNA from multiple organisms may be present and where the aim is not to assemble a single genome, but to understand the wider biological composition of the sample.

The last 2 years has seen the emergence of a number of specialised metagenomic assemblers aimed at addressing this tools gap. One example of this is our own work, MetaCortex, which is capable of assembling sequencing reads and outputting contigs indicative of the entire set of organisms present in a sample. MetaCortex works by identifying sub-graphs within the wider de Bruijn graph assembly and building contigs, which represent the consensus path through these components. In doing so, variation information is preserved and it is possible to detect organisms present in the sample with low read coverage. We applied this method to the analysis of DNA samples from Eidolon helvum, a species of bat implicated in the transmission of zoonotic disease to humans and were able to identify hundreds of different viral DNA signatures within samples.

References Hohage T. and Pricop M."Nonlinear Tikhonov regularization in Hilbert scales for inverse boundary value problems with random noise".Inverse Problems and Imaging, Vol. 2, 271{ 290, 2008. Bissantz N., Hohage T. and Munk A."Consistency and rates of convergence of nonlinear Tikhonov regularization with random noise". Inverse Problems, Vol. 20, 1773{1791, 2004.

Production of biogas by means of anaerobic digestion of biomass is becoming increasingly important as biogas is regarded a clean, renewable and environmentally compatible energy source. Moreover, generation of energy from biogas relies on a balanced carbon dioxide cycle. In 2013, more than 7,500 digesters were operating in Germany, generating 25 TWh of electricity per year. The microbiology of biogas fermentation from biomass is complex and involves interaction of different microorganisms. The majority of the participating microbes are still unknown, as is their influence on reactor performance.

Our study aims at the elucidation of the microbiology of biogas producing microbial communities by applying metagenome, metatranscriptome and single cell sequencing of dominant community members, combined with metaproteomics. In this study we profile four different production scale biogas plants. Extracted whole community DNA and RNA were deeply sequenced using the Illumina’s HiSeq sequencer, while 16S rRNA gene sequencing was done on the MiSeq system. Metagenome sequencing resulted in 403 Gbp total sequence information, while metatranscriptome sequencing yielded 290 Gbp.

The obtained sequencing depth enables a profound taxonomic and functional analysis, as well as the reconstruction of microbial genomes representing so far uncultured species and relevant metabolic pathways. This is the first study of production-scale biogas plant microbial communities combining metagenomics, metatranscriptomics, metaproteomics and single cell genome sequencing, providing comparative insights into the microbiology of different fermenters, with respect to differences in operation conditions and fed substrates, different pathways involved in substrate hydrolysis, acidogenesis, acetogenesis, methane production and stress response.

Soil is the most biodiverse environment on earth, typically containing 109 bacterial cells from 106 different species per g. At most, 1% of these cells can grow in the laboratory; the majority remained obscure until the recent development of culture-independent methods. Usually taken for granted, soil is an invaluable resource providing essential ecosystem services as well as food to sustain the growing human population. More information on the genetic diversity of soil communities, in particular functional genes for the biological processes that underpin soil quality, is needed to establish the resilience of the system to perturbation. How many individuals and phylogenetic groups are potentially capable of performing a function? How many of these are active in particular conditions? Are there multiple possible combinations that provide similar functionality? Does the system always return to one combination after perturbation? This is especially relevant to arable agriculture where soils are deliberately manipulated to support crop rotations, during conversion of land to different management systems or with increasingly unpredictable climate change.

The advent of NGS-based metagenomics and metatranscriptomics with quantitative data on the presence and activity of phylogenetic and functional groups provides an unprecedented opportunity to describe and interpret soil biological systems. Long-term field experiments at Rothamsted Research enable comparison of the relative impact of different fertilizer inputs, crops and cultivation on microbial communities. Data amassed on these soils including 16S rRNA gene amplicons, full metagenomes and transcriptomes will be presented. At present, the large volumes of the datasets, the lack of appropriate analytical pipelines and relevant statistics is a constraint on processing and analysing the data. Nevertheless, it has generated information on which groups respond to nitrogen fertilizer, ploughing and changes in plant cover.

Many metagenomic studies compare hundreds to thousands of environmental and health-related samples by extracting and sequencing their DNA. However, one of the first steps - to determine what bacteria are actually in the sample - can be a computationally time-consuming task since most methods rely on computing the classification of each individual read out of tens to hundreds of thousands of reads. We introduce Quikr: a QUadratic, K-mer based, Iterative, Reconstruction method which computes a vector of taxonomic assignments and their proportions in the sample using an optimization technique motivated from the mathematical theory of compressive sensing. On both simulated and actual biological data, we demonstrate that Quikr is typically more accurate as well as typically orders of magnitude faster than the most commonly utilized taxonomic assignment techniques for both whole genome techniques (Metaphyler, Metaphlan) and 16S rRNA techniques (the Ribosomal Database Project's Naive Bayesian Classifier). We also show that in general nonnegative L1 minimization can be reduced to a simple nonnegative least squares problem.

Related Links: http://www.math.uga.edu/~foucart/WGSQuikr_final.pdf - WGSQuikr preprint

http://www.math.uga.edu/~foucart/QuikrRevisionPrePrint.pdf - Quikr preprint

http://www.math.uga.edu/~foucart/SRasNNLS_final.pdf - Sparse recovery by means of nonnegative least squares

Background: Metagenomics is a rapidly growing field of research that aims at studying assemblages of uncultured organisms with the help of sequencing, with the hope of understanding the true diversity of microbes, their functions, cooperation and evolution. While early papers studied isolated or small numbers of samples, there is now an increasing number of projects that involve systematically collecting multiple samples of, due to sinking sequencing costs, growing size. Moreover, more attention is being paid to the problem of recording relevant environmental parameters (so-called metadata). There is a need for tools that allow one to store and analyze multiple metagenomic datasets in the context of their metadata.

Results: We announce an extension to our metagenome analysis tool MEGAN, called MeganServer, that allows one to store metagenomic datasets on a secure server in order to reduce redundancy and enhancing the ease of sharing large datasets between project members or making the publicly available. The software allows one, additionally, to capture the metadata associated with datasets and then use it to form new composite datasets by combining primary datasets based on the values of their environmental parameters. While the user can analyze any such combined dataset exactly like a primary dataset using MEGAN, internally, a combined dataset refers back to the primary datasets and thus does not duplicate any reads or matches.

Conclusions: With sinking sequencing costs, metagenomic datasets are growing to sizes too large to be stored locally. Installing MeganServer on an computer cluster or using a publicly available instance allows one to store datasets on a server without losing the benefits of using MEGAN locally. Also, combining datasets based on environmental features is an important step in the comparative analysis of metagenome datasets.

The interpretation of metagenomic count data originating from the current generation of DNA sequencing platforms requires special attention. In particular, the per-sample library sizes often vary by orders of magnitude from the same sequencing run, and the counts are overdispersed relative to a simple Poisson model. These challenges can be addressed using an appropriate mixture model that simultaneously accounts for library size differences and biological variability. This approach is already well-characterized and implemented for RNA-Seq data in R packages such as edgeR and DESeq. We use statistical theory, extensive simulations, and empirical data to show that variance stabilizing normalization using a mixture model like the negative binomial is appropriate for microbiome count data. In simulations detecting differential abundance, normalization procedures based on a Gamma-Poisson mixture model provided systematic improvement in performance over crude proportions or rarefied counts -- both of which led to a high rate of false positives. In simulations evaluating clustering accuracy, we found that the rarefying procedure discarded samples that were nevertheless accurately clustered by alternative methods, and that the choice of minimum library size threshold was critical in some settings, but with an optimum that is unknown in practice. Techniques that use variance stabilizing transformations by modeling microbiome count data with a mixture distribution, such as those implemented in edgeR and DESeq, substantially improved upon techniques that attempt t o normalize by rarefying or crude proportions. Based on these results and well-established statistical theory, we advocate that investigators avoid rarefying altogether. We have provided microbiome-specific extensions to these tools in the R package, phyloseq.

Related Links:

http://arxiv.org/abs/1310.0424 - Arxiv Version of Paper.

http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0061217 - Phyloseq Package Description and Philosophy

For many practical applications, for example to uncover the pathogen that caused an infection after the acute phase, very deep short read sequencing can be effective provided that we can reliably assign short sequencing reads to species. This problem of assignment of reads to species is complicated by the fact that, in the absence of very large contigs, most short reads reads match to multiple species. This is essentially a mixture model, where the complete knowledge of all species present in the mixture provides information about the assignment of each read individually. However, metagenomic data analysis rarely formulates the problem in these terms because the very large number of potential species typically makes the inference intractable. Here, we propose a Bayesian model averaging strategy designed to explore the high dimensional space of species present in a metagenomic mixture. We use approximate Bayesian computation and a Monte Carlo strategy to implement the search o f the most appropriate mixture models. Owing to the computationally intensive aspects of the work, we used a population Monte Carlo Markov Chain to leverage the use of parallel computing. We find that the methodolgy is effective to provide a full Bayesian inference for samples with > 10M reads, hence providing interpretable Bayes Factors and posterior probabilities for practical problems that regularly arise in a clinical context.