The past week I have been busy preparing my lectures and practical for the model-based summer school 2017. It was great fun to meet these students and I have learned a lot myself, setting up the practical. Rather than using a dataset of Birte’s lab, this year I used three datasets from openfmri.org, which are already in the BIDS-format. This makes it very easy to use open source tools like fmriprep.
I was very impressed how easy it now is to run very sophisticated preprocessing workflows, using a plethora of neuroimaging software packages and libraries, using the virtual machine approach of Docker and Singularity. On the LISA cluster it was a bit more challenging. Because Singularity automatically mounts your $HOME directory, the packages installed in the container then starts looking for dotfiles (.local/lib/python) it should not use. A solution is to (temprorally) rename your .local-folder, or use something like this, where we overwrite an environment variable (PYTHONUSERBASE) in the virtual environment:
<some_bogus_folder> singularity run -B /run/shm/:/run/shm/ -e poldracklab_fmriprep_latest- 2017-07-20-11e274f76dc3.img /home/gholland/data/bistable /home/gholland/data/bistable/ derivatives participant --participant-label 01
The next thing that was really cool to look into is traditional massively univariate analyses in Python using nistats, a derivative of nilearn, developed in Paris and Berkeley.
Anyway, to see how to do an entire model-based neuroimaging-analysis in just a few dozen lines of code, have a look at the notebooks I prepared for the summer school here: