What it does
LDA Effect Size (LEfSe) (Segata et. al 2010) is an algorithm for high-dimensional biomarker discovery and explanation that identifies genomic features (genes, pathways, or taxa) characterizing the differences between two or more biological conditions (or classes, see figure below). It emphasizes both statistical significance and biological relevance, allowing researchers to identify differentially abundant features that are also consistent with biologically meaningful categories (subclasses). LEfSe first robustly identifies features that are statistically different among biological classes. It then performs additional tests to assess whether these differences are consistent with respect to expected biological behavior.
Specifically, we first use the non-parametric factorial Kruskal-Wallis (KW) sum-rank test to detect features with significant differential abundance with respect to the class of interest; biological significance is subsequently investigated using a set of pairwise tests among subclasses using the (unpaired) Wilcoxon rank-sum test. As a last step, LEfSe uses Linear Discriminant Analysis to estimate the effect size of each differentially abundant feature and, if desired by the investigator, to perform dimension reduction.
LEfSe consists of six modules performing the following steps (see the figure below).
The first step consists of uploading your file by using Galaxy's "Get-Data / Upload-file"
The next steps are:
- A) Format Data for LEfSe: selects the structure of the problem (classes, subclasses, subjects) and formats the tabular abundance data for the B module
- B) LDA Effect Size (LEfSe): performs the analysis using the data formatted with module A and provides input for the visualization modules (C, D, E, F)
- C) Plot LEfSe Results: graphically reports the discovered biomarkes (output of B) with their effect sizes
- D) Plot Cladogram: graphically represents the discovered biomarkers (output of B) in a taxonomic tree specified by the hierarchical feature names (not available for non-hierarchical features)
- E) Plot One Feature: plots the row values of a feature (biomarker or not) as an abundance histogram with classes and subclasses structure (only one feature at the time)
- F) Plot Differential Features: plots the row values of all features (biomarkers or not) as abundance histograms with classes and subclasses structure and provides a zip archive of the figures
Input file format
The text tab-delimited input file consists of a list of numerical features, the class vector and optionally the subclass and subject vectors. The features can be read counts directly or abundance floating-point values more generally, and the first field is the name of the feature. Class, subclass and subject vectors have a name (the first field) and a list of non-numerical strings.
Although both column and row feature organization is accepted, given the high-dimensional nature of metagenomic data, the listing of the features in rows is preferred. A partial example of an input file follows (all values are separated by single-tab):
bodysite mucosal mucosal mucosal mucosal mucosal non_mucosal non_mucosal non_mucosal non_mucosal non_mucosal subsite oral gut oral oral gut skin nasal skin ear nasal id 1023 1023 1672 1876 1672 159005010 1023 1023 1023 1672 Bacteria 0.99999 0.99999 0.999993 0.999989 0.999997 0.999927 0.999977 0.999987 0.999997 0.999993 Bacteria|Actinobacteria 0.311037 0.000864363 0.00446132 0.0312045 0.000773642 0.359354 0.761108 0.603002 0.95913 0.753688 Bacteria|Bacteroidetes 0.0689602 0.804293 0.00983343 0.0303561 0.859838 0.0195298 0.0212741 0.145729 0.0115617 0.0114511 Bacteria|Firmicutes 0.494223 0.173411 0.715345 0.813046 0.124552 0.177961 0.189178 0.188964 0.0226835 0.192665 Bacteria|Proteobacteria 0.0914284 0.0180378 0.265664 0.109549 0.00941215 0.430869 0.0225884 0.0532684 0.00512034 0.0365453 Bacteria|Firmicutes|Clostridia 0.090041 0.170246 0.00483188 0.0465328 0.122702 0.0402301 0.0460614 0.135201 0.0115835 0.0537381
In this case one may want to use bodysite as class, subsite as subclass and id as subject. Notice that the features have a hierarchical structure specified using the character |.
Input file sample
You can try the LEfSe modules using the dataset available here. You can upload the dataset using Galaxy's Get-Data / Upload File
This is a 16S dataset from (Garrett et. al 2010) and (Veiga et. al 2010) for studying the characteristics of the fecal microbiota in a mouse model of spontaneous colitis. The dataset contains 30 abundance profiles (obtained processing the 16S reads with RDP) belonging to 10 rag2 (control) and 20 truc (case) mice. The metadata consists in class information only, as we don't have subject or subclass information. The same dataset is used to show the graphical results in the module descriptions.
STEP A:
What STEP A does
Preprocessing module for the biomarker discovery tool called LEfSe:
This module of LEfSe preprocesses metagenomic abundance data for the analyses to be carried out with the "Run LEfSe" module. This module is separated from the "Run LEfSe" because one may want to preprocess the data only once but run multiple analyses.
For an overview of LEfSe please refer to the "Introduction" module or to (Segata et. al 2011).
Input format
The module accepts tabular data with the feature list in rows or columns.
Output format
The module generates data readable by the "Run LEfSe" module only.
Parameters
The class vector represents the labels of the main condition under investigation. The (optional) subclass vector denotes the internal groupings with biological meaning inside each class (subclasses of different classes with the same name are processed as different subclasses). The subject vector (optional) reports a third dimension denoting meta-data (subject id, sample type, ... ) which is independent from the class and subclass definition.
The labels can have a hierarchical organization (see example below) reflecting taxonomies (like NCBI or RDB microbial taxonomy, SEED subsystems or GO terms). The taxonomic levels are specified using the character |.
The per-sample normalization is usually applied for metagenomic data in which the relative abundances are taken into account.
Example
Although both column and row feature organization is accepted, given the high-dimensional nature of metagenomic data, the listing of the features in rows is preferred. A partial example of an input file follows (all values are separated by single-tab):
bodysite mucosal mucosal mucosal mucosal mucosal non_mucosal non_mucosal non_mucosal non_mucosal non_mucosal subsite oral gut oral oral gut skin nasal skin ear nasal id 1023 1023 1672 1876 1672 159005010 1023 1023 1023 1672 Bacteria 0.99999 0.99999 0.999993 0.999989 0.999997 0.999927 0.999977 0.999987 0.999997 0.999993 Bacteria|Actinobacteria 0.311037 0.000864363 0.00446132 0.0312045 0.000773642 0.359354 0.761108 0.603002 0.95913 0.753688 Bacteria|Bacteroidetes 0.0689602 0.804293 0.00983343 0.0303561 0.859838 0.0195298 0.0212741 0.145729 0.0115617 0.0114511 Bacteria|Firmicutes 0.494223 0.173411 0.715345 0.813046 0.124552 0.177961 0.189178 0.188964 0.0226835 0.192665 Bacteria|Proteobacteria 0.0914284 0.0180378 0.265664 0.109549 0.00941215 0.430869 0.0225884 0.0532684 0.00512034 0.0365453 Bacteria|Firmicutes|Clostridia 0.090041 0.170246 0.00483188 0.0465328 0.122702 0.0402301 0.0460614 0.135201 0.0115835 0.0537381
In this case one may want to use bodysite as class, subsite as subclass and id as subject. Notice that the features have a hierarchical structure specified using the character |.
Example with the "mouse model dataset"
You can try the LEfSe modules using the dataset available here. This is a 16S dataset from (Garrett et. al 2010) and (Veiga et. al 2010) for studying the characteristics of the fecal microbiota in a mouse model of spontaneous colitis. The dataset contains 30 abundance profiles (obtained processing the 16S reads with RDP) belonging to 10 rag2 (control) and 20 truc (case) mice. The metadata consists of class information only, as we don't have subject or subclass information. The dataset contains the features organized in rows; you need to select the first row as class, whereas you have to select "no subclass" and "no subject" options.
How to Cite LEfSe
If you find LEfSe usefull in your research please city our paper (Segata et. al 2010):
Please do not hesitate to contact us for any questions of comments.