compare_categories.py – Analyzes statistical significance of sample groupings using distance matrices

Description:

This script allows for the analysis of the strength and statistical significance of sample groupings using a distance matrix as the primary input. Several statistical methods are available: adonis, ANOSIM, BIO-ENV, Moran’s I, MRPP, PERMANOVA, PERMDISP, and db-RDA.

Note: R’s vegan and ape packages are used to compute many of these methods, and for the ones that are not, their implementations are based on the implementations found in those packages. It is recommended to read through the detailed descriptions provided by the authors (they are not reproduced here) and to refer to the primary literature for complete details, including the methods’ assumptions. To view the documentation of a method in R, prepend a question mark before the method name. For example:

?vegan::adonis

The following are brief descriptions of the available methods:

adonis - Partitions a distance matrix among sources of variation in order to describe the strength and significance that a categorical or continuous variable has in determining variation of distances. This is a nonparametric method and is nearly equivalent to db-RDA (see below) except when distance matrices constructed with semi-metric or non-metric dissimilarities are provided, which may result in negative eigenvalues. adonis is very similar to PERMANOVA, though it is more robust in that it can accept either categorical or continuous variables in the metadata mapping file, while PERMANOVA can only accept categorical variables. See vegan::adonis for more details.

ANOSIM - Tests whether two or more groups of samples are significantly different based on a categorical variable found in the metadata mapping file. You can specify a category in the metadata mapping file to separate samples into groups and then test whether there are significant differences between those groups. For example, you might test whether ‘Control’ samples are significantly different from ‘Fast’ samples. Since ANOSIM is nonparametric, significance is determined through permutations. See vegan::anosim for more details.

BIO-ENV - Finds subsets of variables whose Euclidean distances (after scaling the variables) are maximally rank-correlated with the distance matrix. For example, the distance matrix might contain UniFrac distances between communities, and the variables might be numeric environmental variables (e.g., pH and latitude). Correlation between the community distance matrix and Euclidean environmental distance matrix is computed using Spearman’s rank correlation coefficient (rho). This method will only accept categories that are numerical (continuous or discrete). This is currently the only method in the script that accepts more than one category (via -c). See vegan::bioenv for more details. This method is also known as BEST (previously called BIO-ENV) in the PRIMER-E software package.

Moran’s I - This method uses the numerical (e.g. geographical) data supplied to identify what type of spatial configuration occurs in the samples. For example, are they dispersed, clustered, or of no distinctly noticeable configuration when compared to each other? This method will only accept a category that is numerical. See ape::Moran.I for more details.

MRPP - This method tests whether two or more groups of samples are significantly different based on a categorical variable found in the metadata mapping file. You can specify a category in the metadata mapping file to separate samples into groups and then test whether there are significant differences between those groups. For example, you might test whether ‘Control’ samples are significantly different from ‘Fast’ samples. Since MRPP is nonparametric, significance is determined through permutations. See vegan::mrpp for more details.

PERMANOVA - This method is very similar to adonis except that it only accepts a categorical variable in the metadata mapping file. It uses an ANOVA experimental design and returns a pseudo-F value and a p-value. Since PERMANOVA is nonparametric, significance is determined through permutations.

PERMDISP - This method analyzes the multivariate homogeneity of group dispersions (variances). In essence, it determines whether the variances of groups of samples are significantly different. The results of both parametric and nonparametric significance tests are provided in the output. This method is generally used as a companion to PERMANOVA. See vegan::betadisper for more details.

db-RDA - This method is very similar to adonis and will only differ if certain non-Euclidean semi- or non-metrics are used to generate the input distance matrix, and negative eigenvalues are encountered. The only difference then will be in the p-values, not the R^2 values. As part of the output, an ordination plot is also generated that shows grouping/clustering of samples based on a category in the metadata mapping file. This category is used to explain the variability between samples. Thus, the ordination output of db-RDA is similar to PCoA except that it is constrained, while PCoA is unconstrained (i.e. with db-RDA, you must specify which category should be used to explain the variability in your data). See vegan::capscale for more details.

For more information and examples pertaining to this script, please refer to the accompanying tutorial, which can be found at http://qiime.org/tutorials/category_comparison.html.

Usage: compare_categories.py [options]

Input Arguments:

Note

[REQUIRED]

--method

The statistical method to use. Valid options: adonis, anosim, bioenv, morans_i, mrpp, permanova, permdisp, dbrda

-i, --input_dm

The input distance matrix. WARNING: Only symmetric, hollow distance matrices may be used as input. Asymmetric distance matrices, such as those obtained by the UniFrac Gain metric (i.e. beta_diversity.py -m unifrac_g), should not be used as input

-m, --mapping_file

The metadata mapping file

-c, --categories

A comma-delimited list of categories from the mapping file. Note: all methods except for BIO-ENV accept just a single category. If multiple categories are provided, only the first will be used

-o, --output_dir

Path to the output directory

[OPTIONAL]

-n, --num_permutations

The number of permutations to use when calculating statistical significance. Only applies to adonis, ANOSIM, MRPP, PERMANOVA, PERMDISP, and db-RDA. Must be greater than or equal to zero [default: 999]

Output:

At least one file will be created in the output directory specified by -o. For most methods, a single output file containing the results of the test (e.g. the effect size statistic and p-value) will be created. The format of the output files will vary between methods as some are generated by native QIIME code, while others are generated by R’s vegan or ape packages. Please refer to the script description for details on how to access additional information for these methods, including what information is included in the output files.

db-RDA is the only exception in that two output files are created: a results text file and a PDF of the ordination plot.

adonis example:

Runs the adonis statistical method on a distance matrix and mapping file using the Treatment category and 999 permutations, writing the output to the ‘adonis_out’ directory.

compare_categories.py --method adonis -i unweighted_unifrac_dm.txt -m Fasting_Map.txt -c Treatment -o adonis_out -n 999

ANOSIM example:

Runs the ANOSIM statistical method on a distance matrix and mapping file using the Treatment category and 99 permutations, writing the output to the ‘anosim_out’ directory.

compare_categories.py --method anosim -i unweighted_unifrac_dm.txt -m Fasting_Map.txt -c Treatment -o anosim_out -n 99