Metagenomic biomarker discovery and explanation.
Segata N, Izard J, Waldron L, Gevers D, Miropolsky L, Garrett WS, Huttenhower C. · 2011
Read Metagenomic biomarker discovery and explanation. by Segata N, Izard J, Waldron L, Gevers D, Miropolsky L, Garrett WS, Huttenhower C. (2011) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc3218848.txt.
Introduction
Nicola Segata and colleagues described LEfSe, a method for finding biomarkers in metagenomic data, in this 2011 paper in Genome Biology. Metagenomics profiles whole microbial communities, and a central problem is identifying which organisms, genes, or pathways reliably separate one community type from another. Their method combines several checks into one pipeline.
It first tests for features that differ between classes. Then it checks that the differences are biologically consistent. Finally it estimates the size of each effect. Results are thus ranked by importance, not by statistical significance alone. The authors tested the approach widely on several real microbiomes. They also provided a web interface, so researchers without local computing power could use the tool.
Catalog record
| CREATOR | Segata N, Izard J, Waldron L, Gevers D, Miropolsky L, Garrett WS, Huttenhower C. |
|---|---|
| YEAR | 2011 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 93 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Metagenomic biomarker discovery and explanation By Nicola Segata, Jacques Izard, Levi Waldron, Dirk Gevers, Larisa Miropolsky, Wendy S Garrett, Curtis Huttenhower Published in: Genome Biology DOI: 10.1186/gb-2011-12-6-r60 ABSTRACT Abstract This study describes and validates a new method for metagenomic biomarker discovery by way of class comparison, tests of biological consistency and effect size estimation. This addresses the challenge of finding organisms, genes, or pathways that consistently explain the differences between two or more microbial communities, which is a central problem to the study of metagenomics. We extensively validate our method on several microbiomes and a convenient online interface for the method is provided at http://huttenhower.sph.harvard.edu/lefse/. BACKGROUND Biomarker discovery has proven to be one of the most broadly applicable and successful means of translating molecular and genomic data into clinical practice. Comparisons between healthy and diseased tissues have highlighted the importance of tasks such as class discovery (detecting novel subtypes of a disease) and class prediction (determining the subtype of a new sample) [1-4], and recent metagenomic assays have shown that human microbial communities can be used as biomarkers for host factors such as lifestyle [5-7] and disease [7-10]. As sequencing technology continues to develop and makes microbial biomarkers increasingly easily detected, this enables clinical diagnostic and microbiological applications through the comparison of microbial communities [11,12]. The human microbiome, consisting of the total microbial complement associated with human hosts, is an important emerging area for metagenomic biomarker discovery [13,14]. Changes in microbial abundances in the gut, oral cavity, and skin have been associated with disease states ranging from obesity [15-17] to psoriasis [18]. More generally, the metagenomic study of microbial communities is an effective approach for identifying the microorganisms or microbial metabolic characteristics of any uncultured sample [19,20]. Analyses of metagenomic data typically seek to identify the specific organisms, clades, operational taxonomic units, or pathways whose relative abundances differ between two or more groups of samples, and several features of microbial communities have been proposed as potential biomarkers for various disease states. For example, single pathogenic organisms can signal disease if present in a community [21,22], and increases and decreases in community complexity have been observed in bacterial vaginosis [23] and Crohn's disease [8]. Each of these different types of microbial biomarkers is correlated with disease phenotypes, but few bioinformatic methods exist to explain the class comparisons afforded by metagenomic data. Identifying the most biologically informative features differentiating two or more phenotypes can be challenging in any genomics dataset, and this is particularly true for metagenomic biomarkers. Robust statistical tools are needed to ensure the reproducibility of conclusions drawn from metagenomic data, which is crucial for clinical application of the biological findings. Related challenges are associated with high-dimensional data regardless of the data type or experimental platform; the number of potential biomarkers, for example, is typically much higher than the number of samples [24-26]. Metagenomic analyses additionally present their own specific issues, including sequencing errors, chimeric reads [27,28], and complex underlying biology; many microbial communities have been found to show remarkably high inter-subject variability. For example, large differences are detected even among the gut microbiomes of twins [29], and both human microbiomes and environmental communities are thought to be characterized by the presence of a long tail of rare organisms [30-32]. Moreover, simply identifying potential biomarkers without elucidating their biological consistency and roles is only a precursor to understanding the underlying mechanisms of microbe-microbe or host-microbe interactions [33]. In many cases, it is necessary to explain not just how two biological samples differ, but why. This problem is referred to as class comparison: how can the differences between phenotypes such as tumor subtype or disease state be explained in terms of consistent biological pathways or molecular mechanisms? A number of methods have been proposed for class discovery or comparison in metagenomic data. MEGAN [34] is a metagenomic analysis tool with recent additions for phylogenetic comparisons [35] and statistical analyses [36]. MEGAN, however, can only compare single pairs of metagenomes, as is also the case with STAMP [37], which does introduce a concept of 'biological relevance' in the form of confidence intervals. UniFrac [38] compares sets of metagenomes at a strictly taxonomic level using phylogenetic distance, while MG-RAST [39], ShotgunFunctionalizeR [40], mothur [41], and METAREP [42] all process metagenomic data using standard statistical tests (mainly t-tests with some modifications). Most methods for community analysis from an ecological perspective rely on unsupervised cluster analyses based on principal component analysis [43] or principal coordinate analysis [44]. These can successfully detect groups of related samples, but they fail to include prior knowledge of phenotypes or environmental conditions associated with the groups, and they generally do not identify the biological features responsible for group relationships. Metastats [45] is the only current method that explicitly couples statistical analysis (to assess whether metagenomes differ) with biomarker discovery (to detect features characterizing the differences) based on repeated t statistics and Fisher's tests on random permutations. However, none of these methods, even those offering nuanced analyses of metagenomic data, provide biological class explanations to establish statistical significance, biological consistency, and effect size estimation of predicted biomarkers. In this work, we present the linear discriminant analysis (LDA) effect size (LEfSe) method to support high-dimensional class comparisons with a particular focus on metagenomic analyses. LEfSe determines the features (organisms, clades, operational taxonomic units, genes, or functions) most likely to explain differences between classes by coupling standard tests for statistical significance with additional tests encoding biological consistency and effect relevance. Class comparison methods typically predict biomarkers consisting of features that violate a null hypothesis of no difference between classes; we additionally detect the subset of features with abundance patterns compatible with an algorithmically encoded biological hypothesis and estimate the sizes of the significant variations. In particular, effect size provides an estimation of the magnitude of the observed phenomenon due to each characterizing feature and it is thus a valuable tool for ranking the relevance of different biological aspects and for addressing further investigations and analyses. The introduction of prior biological knowledge in the method contributes to constrain the analysis and thus to address the challenges traditionally connected with high-dimensional data mining. LEfSe thus aims to support biologists by suggesting biomarkers that explain most of the effect differentiating phenotypes of interest (two or more) in biomarker discovery comparative and hypothesis-driven investigations. The visualization of the discovered biomarkers on taxonomic trees provides an effective means for summarizing the results in a biologically meaningful way, as this both statistically and visually captures the hierarchical relationships inherent in 16S-based taxonomies/phylogenies or in ontologies of pathways and biomolecular functions. We validated this approach using data from human microbiomes, a mouse model of ulcerative colitis, and environmental samples, in each case predicting groups of organisms or operational taxonomic units that concisely differentiate the classes being compared. We further evaluated LEfSe using synthetic data, observing that it achieves a substantially better false positive rate compared to standard statistical tests, at the price of a moderately increased false negative rate (that can be adjusted as needed by the user). An implementation of LEfSe including a convenient graphical interface incorporated in the Galaxy framework [46,47] is provided online at [48]. RESULTS AND DISCUSSION LEfSe is an algorithm for high-dimensional biomarker discovery and