sparcc
The sparcc command allows you to .… To run this tutorial please download: fake_data.1.subsample.shared.zip.
Default
The shared parameter is required.
mothur > sparcc(shared=fake_data.1.subsample.shared)
Options
samplings
The samplings parameter is used to .… Default=20.
mothur > sparcc(shared=fake_data.1.subsample.shared, samplings=10)
iterations
The iterations parameter is used to .…Default=10.
mothur > sparcc(shared=fake_data.1.subsample.shared, iterations=100)
permutations
The permutations parameter is used to .…Default=1000.
mothur > sparcc(shared=fake_data.1.subsample.shared, permutations=10000)
method
The method parameter is used to .…Options are relabund and dirichlet. Default=dirichlet.
mothur > sparcc(shared=fake_data.1.subsample.shared, method=relabund)
groups
The groups parameter is used to select groups from your shared file. The default value for groups is all the groups in your shared file. Group names may be separated by dashes.
mothur > sparcc(shared=fake_data.1.subsample.shared, groups=0-1-2)
label
The label parameter is used to analyze specific labels in your shared file.
processors
The processors parameter allows you to run the command with multiple processors. Default processors=Autodetect number of available processors and use all available.
mothur > sparcc(shared=fake_data.1.subsample.shared, processors=2)
Revisions
- 1.31.0 - First Introduced
- 1.39.0 - Fixes floating point exception
- 1.40.0 - Speed and memory improvements for shared files. #357 , #347
- 1.40.0 - Rewrite of threaded code. Default processors=Autodetect number of available processors and use all available.
- 1.40.3 - Improves speed with threaded code and corrects small bug.