Analysis with Nipype

From MEG Core
Jump to navigation Jump to search

UNDER CONSTRUCTION
Nipype is a pipeline configuration tool that controls analysis flow and easily reconfigures for added functionality. Nipye controls the flow of data and automatically determines dependencies to create filenames and folders

Nipype meg.png

Nipype Resources

https://nipype.readthedocs.io/en/latest/
https://miykael.github.io/nipype_tutorial/

Nipype configuration for Biowulf

pull pyctf

 module load fftw/3.3.8/openmpi-4.0.0/gcc-7.3.0  ## it is possible to use another version - do 'module spider fftw'
 cd $PYCTF_DIR  #Folde will have the Makefile
 make clean
 make
 make install
 make usersite  (will add a link so package is seen in conda)

pull SAMsrcV5

 For installation instructions read the INSTALL text file - somewhat summarized below
 cd $SAMSRCV5_DIR
 make
 cd config
 make symlinks

pull megblocks
install miniconda (or Anaconda)

 https://hpc.nih.gov/apps/python.html#envs   << Go to the Creating Python Environments Section to configure
 *Add environment information to miniconda   ##Need to provide environment.yml
  • Add pyctf, SAMsrc, and megblocks to the miniconda environment ## TODO < provide a way to do this. Add the paths to miniconda/env......pth file


Running Pipeline

Login to biowulf

 # --cpus-per-task=Multithreading amount
 # --ntasks=Number of Parrallel tasks
 # Total CPU count = (cpus-per-task * ntasks)  - Most nodes support 56 max CPUs
 sinteractive --exclusive --cpus-per-task=8 --mem-per-cpu=2g --ntasks=6
 export OMP_NUM_THREADS=4  #Otherwise these tasks will try to use all of the cores
 module load afni
 module load ctf
 conda activate mne  #This is whatever environment configured for nipype
 ~/



Nipype has strict requirements on inputs and outputs

  1. InputSpec
  2. OutputSpec



Preconfigured pipeline templates:

  1. Epilepsy
    1. Sam_epi
  2. Task
    1. Parsemarks, addmarks, samcov, sam3d, ... stats?
  3. Resting State


SAM input base class used for all of the SAM functions


Configuration on Biowulf:

  1. Set up conda: instructions on NIH HPC - https://hpc.nih.gov/apps/python.html
  2. Activate the conda environment and install nipype
 #If you do not know your environment name type: conda env list
 conda activate $ENVIRONMENT_NAME   
 conda install -c nipype conda-forge