The Genboree Microbiome Toolset and the analysis of 16S rRNA microbial sequences
- Kevin Riehle†1,
- Cristian Coarfa†1,
- Andrew Jackson†1,
- Jun Ma2,
- Arpit Tandon1,
- Sameer Paithankar1,
- Sriram Raghuraman1,
- Toni-Ann Mistretta3,
- Delphine Saulnier4,
- Sabeen Raza3,
- Maria Alejandra Diaz3,
- Robert Shulman5,
- Kjersti Aagaard2,
- James Versalovic3 and
- Aleksandar Milosavljevic1Email author
© Riehle et al; licensee BioMed Central Ltd. 2012
Published: 24 August 2012
Microbial metagenomic analyses rely on an increasing number of publicly available tools. Installation, integration, and maintenance of the tools poses significant burden on many researchers and creates a barrier to adoption of microbiome analysis, particularly in translational settings.
To address this need we have integrated a rich collection of microbiome analysis tools into the Genboree Microbiome Toolset and exposed them to the scientific community using the Software-as-a-Service model via the Genboree Workbench. The Genboree Microbiome Toolset provides an interactive environment for users at all bioinformatic experience levels in which to conduct microbiome analysis. The Toolset drives hypothesis generation by providing a wide range of analyses including alpha diversity and beta diversity, phylogenetic profiling, supervised machine learning, and feature selection.
We validate the Toolset in two studies of the gut microbiota, one involving obese and lean twins, and the other involving children suffering from the irritable bowel syndrome.
By lowering the barrier to performing a comprehensive set of microbiome analyses, the Toolset empowers investigators to translate high-volume sequencing data into valuable biomedical discoveries.
The Human Microbiome Project (HMP) aims to improve the understanding of the microbiome and the factors that influence the distribution and evolution of constituent microorganisms in a healthy human population cohort. A number of focused sub-projects within HMP aim to detect and interpret perturbations of microbiomes associated with human diseases . These efforts are being been aided by accelerating technical and methodological advancements in sequencing and computational technologies. The 16S rRNA gene has proven to be a useful initial genomic target to identify and differentiate distinct microbial profiles, such as those in human fecal samples . Determining the abundance (and inferred function) of each type of microbe (community profiling) is less expensive using 16S rRNA than bacterial genomic DNA because only one representative gene from each genome is examined . As the focus widens from 16S rRNA to genomic sequencing, as the costs of sequencing decrease, and the amounts of publically available data increase, the technological and methodological bottleneck on the road to discoveries will shift from sequencing to bioinformatic analysis.
The new bioinformatic bottleneck will need to be addressed in an innovative way, particularly with regard to translational research. In the field of metagenomics, the productivity of translational research is increasingly determined by the amount of effort required to integrate large volumes of “omics” data with clinical metadata and analyze the integrated data sets using latest tools to generate biomedically relevant testable hypotheses. There are multiple mature, open source tools for 16S rRNA gene analyses, that are well maintained and widely used within the scientific community, such as QIIME (Quantitative Insights Into Microbial Ecology)  and mothur . We have integrated those tools within the Genboree Microbiome Toolset and deployed them through the Genboree Workbench  using the Software-as-a-Service model. To enable researchers to gain insight into clinically relevant phenotypes, while accounting for the most significant confounding factors, we have designed the Genboree Microbiome Toolset to be “sample centric”. The Toolset enables users to associate metadata with samples for both supervised sample classification and unsupervised analyses. The toolset also enables analyses of alpha diversity and beta diversity, phylogenetic analysis, and feature selection. The Toolset assures reproducibility of results reported in journal publications and provides default settings at each step that can be customized by the user to reflect their preference or to follow a standard protocol.
The Toolset is deployed using the easy-to-use web-based GUI environment of the Genboree Workbench. The Genboree environment enables web-based collaboration while allowing access control to sensitive data. By virtue of integration through the Genboree Workbench, all the functionality that is accessible interactively through the Toolset is also accessible programmatically via a custom REST Application Programming Interface [7, 8], the Genboree REST API , thus allowing programmatic extension, customization, and web-based integration.
Linking quality filtered sequences to sample metadata
Massively parallel sequencing platforms such as 454 typically produce multiplexed sequence files that contain sequences from more than one sample. Samples and their associated metadata are linked to corresponding sequences using the MID (multiplex identifier), proximal primer, and distal primer. For the purpose of downstream analyses, samples may be analyzed individually or as sample sets.
Taxonomic classification via the Ribosomal Database Project (RDP) Classifier
The Microbiome Toolset integrates the Ribosomal Database Project (RDP) Classifier , which performs taxonomic classification of individual 16S rRNA sequences based on a naive Bayesian classification. The output of RDP Classifier 2.1 (and newer) assigns each sequence to the most specific taxon level (from the Domain to the Genus levels). Sequence counts are then calculated for distinct taxa at each of the levels and combined to produce absolute and relative abundance profiles at each level.
Creating the Operational Taxonomic Unit (OTU) table and representative sequences for phylogenetic tree reconstruction
The Microbiome Toolset integrates a range of analyses based on OTUs, groups of sequences distinguished by their mutual similarity. The QIIME package  performs multi-step chained OTU picking using multiple third party tools, including cd-hit , mothur , and uclust . High speed is achieved by using a rough, fast method to collapse sequence groups that have a high level of similarity, followed by a more computationally demanding and rigorous OTU picking step. Chimeric sequences, which can be falsely detected as novel organisms, resulting in the artificial inflation of diversity are detected and removed using Chimera Slayer . A set of sequences representing each OTU are used for phylogenetic tree reconstruction. Sequence counts per OTU and per sample are summarized in an output OTU table, which is a key input for downstream analyses.
The Toolset enables comparison and visualization of representative sequences in the context of a phylogenetic tree. For comparison, we use UniFrac to examine differences between microbiome communities by measuring the distance between sample-specific sets of taxa in a phylogenetic tree. The phylogenetic distances estimate the degree of evolutionary divergence between different representative sequences , not just the degree of their sequence-level differences.
Phylogenetic differences may be visualized using tools such as the interactive Tree Of Life (iTOL), which supports upload, display, and manipulation of phylogenetic trees . The Microbiome Toolset automatically generates a multi-level circular phylogenetic tree based on metadata and taxonomic information via iTOL’s API. Sample-associated metadata can be used to visually detect phylogenetic distribution biases in specific samples or sample sets.
Beta diversity analysis
Beta diversity analysis considers biodiversity between groups of samples, focusing on the elements that are either unique to or shared among groups of samples. The Toolset includes the QIIME pipeline, which currently includes 14 non-phylogenetic, 9 binary non-phylogenetic, and 6 phylogenetic metrics. Beta diversity plots are generated for all the metrics so that the user can explore the differences between various microbiomes. This approach was adopted to increase the likelihood of detecting biologically significant patterns visible only when using specific metrics. A case in point is the Canberra metric, an equal-weight metric that standardizes the input such that each OTU affects the distance value equally . This equal-weight, non-phylogenetic method was successfully used to distinguish between two tundra communities which could not be distinguished using chord distance, an alternative non-phylogenetic method biased towards largely abundant taxa .
The phylogenetic-based UniFrac  algorithm enables the analysis of different microbiomes by providing both a quantitative measurement, using weighted UniFrac, and a qualitative measurement, using unweighted UniFrac. Quantitative measures of a phylogenetic method shows changes in OTU abundance such as those caused by nutrient shifts whereas qualitative measures detect differences in microbiomes based on presence or absence of species in specific environments such as high- or low-temperature .
For each beta diversity distance metric that is utilized, the results are displayed for the top three principal coordinates using Principal Coordinates Analysis (PCoA) for both normalized and non-normalized OTU tables. Normalizing the OTU tables on a sample-by-sample basis allows the researcher to account for potential variability in sequencing depth. PCoA plots in both 2D and 3D formats are provided in embedded HTML for further analysis. Beta diversity clustering has been utilized to show that three different individuals can be discriminated based on their distinct skin (fingertip) microbiomes obtained from their keyboards .
Classification and selection of discriminating features
It is frequently of interest to identify a small and assayable set of OTUs that can distinguish between sets of samples with different phenotypes. To meet this goal, a supervised machine learning pipeline was developed and exposed within the Toolset. The pipeline determines the success rate of classifying groups of samples and selects the features that best discriminate groups of samples. The pipeline utilizes the R package randomForest  for supervised learning and Boruta  for feature selection. The input to the pipeline consists of the OTU table from QIIME pipeline and the sample metadata collected using the Sample Importer.
The algorithm randomForest employs an ensemble approach based on the Classification and Regression Trees (CART) method. It generates and evaluates many classification trees for discrete data or regression trees for continuous data. The classification error rate is measured by the out-of-bag (OOB) error estimation for each metadata category. Because randomForest does not inherently provide for feature selection , we employed the R package Boruta, a feature selection algorithm built around the randomForest algorithm. The Z score, computed by dividing the average loss of accuracy by its standard deviation, associates an importance measure with the randomForest results. In addition to the output files generated by the randomForest and Boruta packages, the Toolset provides a summary file that combines the OTU number, taxonomic labels generated by RDP, metrics of OTU distribution for each metadata group (min, max, and quartiles), Mann-Whitney  U and Z scores, box plot coordinates, and directional change (calculated by comparing Mann-Whitney U scores for each sample group).
The Toolset also includes an R script to visualize the top performing features in a box plot format that mimics the plot from Qin and Li’s study of the human gut microbial gene catalogue . Such plots provide concise and informative visual summaries of directional change and relative abundance for the most discriminating features.
Integration of the Microbiome Toolset within the Genboree Workbench
The Genboree Workbench allows users to navigate through a plethora of data sources and match the data to available tools. The interface exposes various data sources to a user via a folder in the left pane (Fig. 2B-i). The first level is that of user Groups corresponding to a specific collaboration or more permanent groupings such as a specific research laboratory. The next level in the folder system contains Projects and Databases, which further encompass Annotation Objects, Annotation Tracks, Samples, Queries, and unstructured Files. Additional information about a selected object appears in the details panel (Fig. 2B-ii). A user can also use that area to download data items onto a local computer.
The tool interface is drag-and-drop: tool inputs are dragged from the folder system into the Input Data Panel (Fig. 2B-iii) and the target databases are dragged from the folder system into the Output Targets panel (Fig. 2B-iv). Each tool has specific input and output requirements. If these requirements are met, the tool is highlighted in green in the menu and can be invoked by selecting the tool from the menu. If the tool is not highlighted in green, a click on the tool in the menu displays a help dialog which will list the input and output requirements and other useful infomation.
The Microbiome Toolset is founded on a sample-centric data and analysis model. Prior to performing analysis steps using the Toolset, as illustrated in Fig. 1., the users must establish links between input sequences and the samples from which the sequences were derived. The Microbiome Toolset and all other Genboree Workbench tools are invoked with customizable default settings. The analysis steps illustrated in Fig. 1 are seamlessly integrated in the Microbiome Toolset pipeline and require no file format conversions.
The inputs and outputs of specific analysis steps are stored in Genboree Databases and Project pages. Genboree Project pages are automatically generated by the tools but may also be manually edited. Serving a role similar to the role of pages in a paper-based lab-book, Genboree Project pages include links and summaries of previously run data sets, which include links to full results, such as 2D and 3D beta diversity plots, groups of alpha diversity plots, classification rates or relevant OTUs for classification.
We present two representative studies carried out using the Genboree Microbiome Toolset. The two studies in combination exercise all the steps outlined in Fig. 1. The first study describes how a Microbiome Toolset user may reproduce a previously reported analysis of alpha diversity in obese and lean twins  by carrying out analysis on publicly available data. The second study describes how the Toolset was used in a recently published study of the gut microbiota of children suffering from irritable bowel syndrome .
Alpha diversity analysis using publicly available data
The Microbiome Toolset makes it easier to reproduce published results from publicly available data and to make new discoveries by performing meta-analyses of integrated data sets. As a proof of this capability, we set out to analyze data from a recently reported study of core gut microbiomes in obese and lean twins .
Our search of the Sequence Read Archive at NCBI using the query phrase ‘core gut microbiome in obese and lean twins’ yielded data associated with the twin study project SRP000319. We downloaded the experimental data for the V6 16S rRNA primer region (SRX001445), which contains 4 runs, 1.6M spots, and 205.6M bases. We were unable to obtain the MIDs from the SRA experiment XML. We were therefore precluded from de-multiplexing the original SFF files by the regular method, but we were able to find a work-around to solve this problem using de-multiplexed sequence data available on a supplemental data page from the Gordon Lab . The metadata obtained from this exercise was compiled for use on the Microbiome Toolset.
A study of microbiota in children with irritable bowel syndrome
Prior to this study , perturbations of the intestinal microbiota between healthy children and children with IBS were not well defined. The study therefore aimed to examine if any such perturbations could be detected. The gastrointestinal microbiota was analyzed in 22 children with IBS (69 samples) and 22 healthy children (71 samples) ages 7-12 for a total of 140 samples. The samples were analyzed for taxonomic abundance, beta diversity, phylogenetic analysis, and classification by supervised machine learning, as summarized in Fig. 1 and described in previous sections.
Taxonomic abundance using the RDP pipeline
The phylogenetic tree (Fig. 6) does not visually reveal differences in phylogeny by the grouping of colors based on health within the inner ring, but it does shed some light on the phylogenetic relationship of the combined pediatric stool microbiome in terms of taxonomic membership (i.e. ratio of Bacteroidetes to Firmicutes) on the outer ring.
Beta diversity analysis
Classification using supervised machine learning and feature selection
RandomForest classification of the IBS-C and IBS-U subtypes was achieved with up to 98.5% success rate. The IBS-U group was distinguished by the presence and relative abundance of 70 OTUs, whereas the IBS-C group was identified by the presence and relative abundance of 54 OTUs (data not shown). Most of the OTUs that facilitated the classifications of these two IBS subtypes belong to the genera or groups such as Bacteroides, Ruminococcus, Lachnospiraceae Incertae Sedis, Veillonella, and Erysipelotrichaceae. We did not observe any extreme changes of relative abundances from these species groups, suggesting that the aggregate collections of species or strains (not individual species or strains) are likely the source of the high degree of classification.
As demonstrated by the two key use case scenarios, the Genboree Microbiome Toolset provides valuable insights into microbiome diversity and identifies disease-associated operational taxonomic units. The Toolset empowers investigators (notably translational researchers) to carry out a wide range of analyses including alpha diversity and beta diversity, phylogenetic profile analysis, classification by supervised machine learning, and feature selection. By lowering the barrier to performing a comprehensive set of microbiome analyses, the Toolset enables characterization of microbiomes and the discovery of disease-associated perturbations.
The Toolset is exposed using the Software-as-a-Service model via the Genboree Workbench. In addition to interactive use, all the functionality of the Workbench is also available programmatically using the Genboree REST Application Programming Interfaces (REST APIs) for web-based integration into project-specific pipelines. The Microbiome Toolset therefore provides a web-based programming environment for bioinformaticians in which to conduct more advanced or custom microbiome analyses.
We foresee the Microbiome Integrated Toolset evolving in multiple directions. First, based on user feedback and progress in the field, we plan to extend and add new pipelines for 16S rRNA genic analysis. Second, the toolset will be extended to enable analyses based on whole-metagenome sequencing. To achieve this aim and accommodate rapidly increasing sequencing volumes, the Genboree Workbench is designed to seamlessly access cloud computing resources across the web.
Software availability and requirements
The Genboree Microbiome Toolset is part of the Genboree Workbench and can be accessed at the address http://genboree.org/java-bin/workbench.jsp. Supported browsers are Internet Explorer versions 8 and above, Mozilla Firefox versions 7 and above. A tutorial for the Genboree Microbiome Toolset is available as Additional File 1. Additional information can be found at the address http://genboree.org/microbiome.
This project was supported in part by the NIH grant R01HG004009 to AM, NIH grants UH2 DK083990 and UH3 DK083990 to JV, and Burroughs Welcome Fund Preterm Birth Initiative and NIH DP21DP2OD001500-01 grants to KA.
This article has been published as part of BMC Bioinformatics Volume 13 Supplement 13, 2012: Selected articles from The 8th Annual Biotechnology and Bioinformatics Symposium (BIOT-2011). The full contents of the supplement are available online at http://www.biomedcentral.com/1471-2105/13/S13/S1
- Turnbaugh PJ, Ley RE, Hamady M, Fraser-Liggett CM, Knight R, Gordon JI: The human microbiome project. Nature 2007, 449: 804–810. 10.1038/nature06244PubMed CentralView ArticlePubMedGoogle Scholar
- Zoetendal EG, Akkermans AD, De Vos WM: Temperature gradient gel electrophoresis analysis of 16S rRNA from human fecal samples reveals stable and host-specific communities of active bacteria. Applied and environmental microbiology 1998, 64: 3854–3859.PubMed CentralPubMedGoogle Scholar
- Hamady M, Knight R: Microbial community profiling for human microbiome projects: Tools, techniques, and challenges. Genome research 2009, 19: 1141–1152. 10.1101/gr.085464.108PubMed CentralView ArticlePubMedGoogle Scholar
- Caporaso G, Kuczynski J, Stombaugh J, Bittinger K, Bushman F, Costello E, Fierer N, Pena A, Goodrich J, Gordon J, et al.: QIIME allows analysis of high-throughput community sequencing data. Nature Methods 2010, 7: 335–336. 10.1038/nmeth.f.303PubMed CentralView ArticlePubMedGoogle Scholar
- Schloss PD, Westcott SL, Ryabin T, Hall JR, Hartmann M, Hollister EB, Lesniewski RA, Oakley BB, Parks DH, Robinson CJ, et al.: Introducing mothur: open-source, platform-independent, community-supported software for describing and comparing microbial communities. Applied and environmental microbiology 2009, 75: 7537–7541. 10.1128/AEM.01541-09PubMed CentralView ArticlePubMedGoogle Scholar
- Fielding RT: Architectural styles and the design of network-based software architectures. In Thesis (Ph D , Information and Computer Science). Volume 2000. University of California, Irvine; 2000.Google Scholar
- Richardson L, Ruby S: RESTful web services. O'Reilly Media 2007.Google Scholar
- Whittaker RH: Evolution and measurement of species diversity. Taxon 1972, 213–251.Google Scholar
- Wang Q, Garrity GM, Tiedje JM, Cole JR: Naive Bayesian classifier for rapid assignment of rRNA sequences into the new bacterial taxonomy. Applied and environmental microbiology 2007, 73: 5261–5267. 10.1128/AEM.00062-07PubMed CentralView ArticlePubMedGoogle Scholar
- Li W, Godzik A: Cd-hit: a fast program for clustering and comparing large sets of protein or nucleotide sequences. Bioinformatics 2006, 22: 1658–1659. 10.1093/bioinformatics/btl158View ArticlePubMedGoogle Scholar
- Edgar RC: Search and clustering orders of magnitude faster than BLAST. Bioinformatics 2010, 26: 2460–2461. 10.1093/bioinformatics/btq461View ArticlePubMedGoogle Scholar
- Haas BJ, Gevers D, Earl AM, Feldgarden M, Ward DV, Giannoukos G, Ciulla D, Tabbaa D, Highlander SK, Sodergren E, et al.: Chimeric 16S rRNA sequence formation and detection in Sanger and 454-pyrosequenced PCR amplicons. Genome Research 2011, 21: 494–504. 10.1101/gr.112730.110PubMed CentralView ArticlePubMedGoogle Scholar
- Lozupone C, Knight R: UniFrac: a new phylogenetic method for comparing microbial communities. Applied and environmental microbiology 2005, 71: 8228–8235. 10.1128/AEM.71.12.8228-8235.2005PubMed CentralView ArticlePubMedGoogle Scholar
- Letunic I, Bork P: Interactive Tree Of Life (iTOL): an online tool for phylogenetic tree display and annotation. Bioinformatics 2007, 23: 127–128. 10.1093/bioinformatics/btl529View ArticlePubMedGoogle Scholar
- Gavin DG, Oswald WW, Wahl ER, Williams JW: A statistical approach to evaluating distance metrics and analog assignments for pollen records. Quaternary Research 2003, 60: 356–367. 10.1016/S0033-5894(03)00088-7View ArticleGoogle Scholar
- Oswald WW, Brubaker LB, Hu FS, Gavin DG: Pollen-Vegetation Calibration for Tundra Communities in the Arctic Foothills, Northern Alaska. Journal of Ecology 2003, 91: 1022–1033. 10.1046/j.1365-2745.2003.00823.xView ArticleGoogle Scholar
- Lozupone CA, Hamady M, Kelley ST, Knight R: Quantitative and qualitative beta diversity measures lead to different insights into factors that structure microbial communities. Applied and environmental microbiology 2007, 73: 1576–1585. 10.1128/AEM.01996-06PubMed CentralView ArticlePubMedGoogle Scholar
- Fierer N, Lauber CL, Zhou N, McDonald D, Costello EK, Knight R: Forensic identification using skin bacterial communities. Proceedings of the National Academy of Sciences of the United States of America 2010, 107: 6477–6481. 10.1073/pnas.1000162107PubMed CentralView ArticlePubMedGoogle Scholar
- Liaw A, Wiener M: Classification and Regression by randomForest. R news 2002, 2: 18–22.Google Scholar
- Kursa MB, Rudnicki WR: Feature selection with the Boruta package.Google Scholar
- Knights D, Costello EK, Knight R: Supervised classification of human microbiota. FEMS Microbiology Reviews 2010, no-no.Google Scholar
- Wilcoxon F: Individual comparisons by ranking methods. Biometrics Bulletin 1945, 1: 80–83. 10.2307/3001968View ArticleGoogle Scholar
- Qin J, Li R, Raes J, Arumugam M, Burgdorf K, Manichanh C, Nielsen T, Pons N, Levenez F, Yamada T, et al.: A human gut microbial gene catalogue established by metagenomic sequencing. Nature 2010, 464: 59–65. 10.1038/nature08821PubMed CentralView ArticlePubMedGoogle Scholar
- Turnbaugh PJ, Hamady M, Yatsunenko T, Cantarel BL, Duncan A, Ley RE, Sogin ML, Jones WJ, Roe BA, Affourtit JP, et al.: A core gut microbiome in obese and lean twins. Nature 2009, 457: 480–484. 10.1038/nature07540PubMed CentralView ArticlePubMedGoogle Scholar
- Saulnier DM, Riehle K, Mistretta TA, Diaz MA, Mandal D, Raza S, Weidler EM, Qin X, Coarfa C, Milosavljevic A, et al.: Gastrointestinal Microbiome Signatures Of Pediatric Patients With Irritable Bowel Syndrome. Gastroenterology 2011.Google Scholar
- The Gordon Lab[http://gordonlab.wustl.edu/SuppData.html]
- Li M, Wang B, Zhang M, Rantalainen M, Wang S, Zhou H, Zhang Y, Shen J, Pang X, Wei H, et al.: Symbiotic gut microbes modulate human metabolic phenotypes. Proceedings of the National Academy of Sciences of the United States of America 2008, 105: 2117–2122. 10.1073/pnas.0712038105PubMed CentralView ArticlePubMedGoogle Scholar
- Magurran AE: Measuring biological diversity. Wiley-Blackwell; 2004.Google Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.