libshuff

The libshuff command implements the libshuff method as previously implemented in the programs s-libshuff and libshuff. The libshuff method is a generic test that describes whether two or more communities have the same structure using the Cramer-von Mises test statistic. The significance of the test statistic indicates the probability that the communities have the same structure by chance. Because each pairwise comparison requires two significance tests, a correction for multiple comparisons (e.g. Bonferroni’s correction) must be applied. The files that we discuss in this tutorial can be obtained by downloading the amazondata.zip file and decompressing it.

Default settings

The group file is a tab-deliminated file where the first column contains the name of the sequence and the second column contains the name of the group that the sequence belongs to. To start this tutorial enter the following command:

mothur > libshuff(phylip=98_lt_phylip_amazon.dist, group=amazon.groups)

By default, the libshuff() command will carry out the integral form of the Cramer-von Mises statistic for each pair of communities using 10,000 randomizations:

This will produce:

********************#****#****#****#****#****#****#****#****#****#****#
forest-pasture      |||||||||||||||||||||||||||||||||||||||||||||||||||
***********************************************************************

Comparison             dCXYScore   Significance
forest-pasture         0.00500678  0.0081
pasture-forest         0.00289973  0.0614

This means that the test statistic for comparing the forest sequences to the pasture sequences is 0.00500678 and for the pasture vs. forest comparison the statistic is 0.00289973. The significance of these values is 0.0081 and 0.0614, respectively. If one desires an experiment-wide false detection rate of 0.05, then these significance values need to be less than 0.025 to be considered statistically significant. If either of the significance values are statistically significant, one can safely state that the two communities are significantly different. Therefore, because 0.0081 is smaller than 0.025, the forest and pasture sequence collections are significantly different.

The libshuff() command will generate two files: 98_lt_phylip_amazon.libshuff.summary and 98_lt_phylip_amazon.libshuff.coverage. The *.summary file contains the same output that is outputted to the screen. The *.coverage file provides the Good’s coverage value within a community and between communities for each distance level.

Options

iters

If you run the libshuff() command multiple times, you will notice that while the significance score doesn’t change, it’s significance may change some. This is because the testing procedure is based on a randomization process that becomes more accurate as you increase the number of randomizations. By default, libshuff() will do 10,000 randomizations. You can change the number of iterations with the iters option as follows:

mothur > libshuff(phylip=98_lt_phylip_amazon.dist, group=amazon.groups, iters=1000)

form

There are two forms of the libshuff method - the integral and discrete forms. The discrete form was originally proposed by Singleton et al. (2001; Applied and Environmental Microbiology) and the more precise integral form was proposed by Schloss et al. (2004; Applied and Environmental Microbiology). By default the libshuff() command implements the integral form of the statistic:

mothur > libshuff(phylip=98_lt_phylip_amazon.dist, group=amazon.groups, form=integral)

Alternatively the discrete form of the statistic can be used:

mothur > libshuff(phylip=98_lt_phylip_amazon.dist, group=amazon.groups, form=discrete)

The default step size used in the default form of the statistic is 0.01.

step

If you use the discrete form, you can set the step size to any value you desire:

mothur > libshuff(phylip=98_lt_phylip_amazon.dist, group=amazon.groups, form=discrete, step=0.001)

If the step size is small enough, it is possible for the discrete and integral forms of the method to yield the same statistic value; however, it will be considerably slower than the integral form.

mothur > libshuff(phylip=98_lt_phylip_amazon.dist, group=amazon.groups, form=discrete, step=0.000001) 

Comparison             dCXYScore   Significance
forest-pasture         0.00500679  0.0093
pasture-forest         0.00289971  0.0599

cutoff

By default the discrete form will consider distances between 0 and 1. To change the upper bound of the analysis, a cutoff can be set using the cutoff option:

mothur > libshuff(phylip=98_lt_phylip_amazon.dist, group=amazon.groups, form=discrete, cutoff=0.10)

Comparison             dCXYScore   Significance
forest-pasture         0.00161599  0.0696
pasture-forest         0.00149521  0.0854