[pymvpa] FDR adjustment after permutation testing

[Nick Oosterhof]

On 18 Jun 2015, at 14:10, H.Y. Chan <chan at rsm.nl> wrote:

> Maybe it's not related to the toolbox, but I'd be grateful if there's anyone who might have advice for me on multiple comparisons.
> 
> I've done permutation testing using the toolbox and got a whole brain map of p values. I wonder if the only FDR procedure possible is voxel-wise?

I consider FDR as pretty evil for whole-brain multiple comparison correction. A strong cluster in one part of the brain can make another, very weak cluster elsewhere, survive.

> Is it possible to do cluster-wise FDR or threshold-free cluster enhancement (TCFE) (I don't know how FSL implements it)? 

PyMVPA supports fixed-threshold Monte Carlo cluster-based correction in mvpa2/algorithms/group_clusterthr.py [1] for volumetric data.

If you are comfortable using Matlab / GNU Octave, consider CoSMoMPVA’s cosmo_montecarlo_cluster_stat [2,3] function, which supports both fixed-threshold and threshold-free cluster enhancement (TFCE) for volumetric and surface-based data.

[1] https://github.com/PyMVPA/PyMVPA/blob/master/mvpa2/algorithms/group_clusterthr.py
[2] https://github.com/CoSMoMVPA/CoSMoMVPA/blob/master/mvpa/cosmo_montecarlo_cluster_stat.m
[3] http://cosmomvpa.org/contents_demo.html#demo-surface-tfce