5.1. Code-less Configuration#

One of the most important features of junifer is its capacity to run without writing a single line of code. This is achieved by using a configuration file that is written in YAML. In this file, we configure the different steps of The junifer Pipeline.

As a reminder, this is how the pipeline looks like:

flowchart LR dg[Data Grabber] dr[Data Reader] pp[Preprocess] mc[Marker Computation] st[Storage] dg --> dr dr --> pp pp --> mc mc --> st

Thus, the configuration file must configure each of the sections of the pipeline, as well as some general parameters.

As an example, we will generate the configuration file for a pipeline that will extract the mean VBM_GM values using two different parcellations and one set of coordinates, from the Oasis VBM Testing dataset included in junifer.

5.1.1. General Parameters#

The general parameters are the ones that are not specific to any of the sections of the pipeline, but configure junifer as a whole. These parameters are:

  • with: A section used to specify modules and junifer extensions to use.

  • workdir: The working directory where junifer will store temporary files.

Since the example uses a specific DataGrabber for testing, we need to add junifer.testing.registry to the with section. This will allow junifer to find the DataGrabber. We will set the workdir to /tmp.

with: junifer.testing.registry

workdir: /tmp

5.1.2. Step-by-step Configuration#

In order to configure the pipeline, we need to configure each step:

  • datagrabber

  • datareader

  • preprocess

  • markers

  • storage

Important

The datareader step configuration is optional, as junifer only provides one DataReader. Nevertheless, it is possible to extend junifer with custom DataReaders, and thus, it is also possible to configure this step.

Data Grabber#

The datagrabber section must be configured using the kind key to specify the DataGrabber to use. Additional keys correspond to the parameters of the DataGrabber constructor.

For example, to use the DataladAOMICPIOP1 DataGrabber, we just need to specify its name as the kind key.

datagrabber:
  kind: DataladAOMICPIOP1

However, it is also possible to pass parameters to the DataGrabber constructor. In this case, we can restrict the DataGrabber to fetch only the restingstate task.

datagrabber:
  kind: DataladAOMICPIOP1
  tasks: restingstate

In the Oasis VBM Testing dataset example, the section will look like this:

datagrabber:
  kind: OasisVBMTestingDataGrabber

Data Reader#

As mentioned before, this section is entirely optional, as junifer only provides one DataReader (DefaultDataReader), which is the default in case the section is not specified.

In any case, the syntax of the section is the same as for the datagrabber section, using the kind key to specify the DataReader to use, and additional keys to pass parameters to the DataReader constructor:

datareader:
  kind: DefaultDataReader

For the Oasis VBM Testing dataset example, we will not specify a datareader step.

Preprocess#

preprocess is also an optional step, as it might be the case that no pre-processing is needed. As we can perform multiple preprocessing steps, it’s passed as a list of Preprocessors. In the case that pre-processing is needed, each Preprocessord must be configured using the kind key to specify the Preprocessor to use, and additional keys to pass parameters to the Preprocessor.

For example, to use the fMRIPrepConfoundRemover Preprocessor, we just need to specify its name as the kind key, as well as its parameters.

preprocess:
  - kind: fMRIPrepConfoundRemover
    strategy:
      motion: full
      wm_csf: full
      global_signal: basic
    spike: 0.2
    detrend: false
    standardize: true

For the Oasis VBM Testing dataset example, we will not specify a preprocessing step.

Marker#

The markers section like the preprocess section expects a list of markers. Each Marker has a name that we can use to refer to it later, and a set of parameters that will be passed to the Marker.

For the Oasis VBM Testing dataset example, we want to compute the mean VBM_GM value for each parcel using the Schaefer parcellation (100 parcels, 7 networks), Schaefer parcellation (200 parcels, 7 networks), and the DMNBuckner network, using 5mm spheres. Thus, we will configure the markers section as follows:

markers:
  - name: Schaefer100x7_mean
    kind: ParcelAggregation
    parcellation: Schaefer100x7
    method: mean
  - name: Schaefer200x7_mean
    kind: ParcelAggregation
    parcellation: Schaefer200x7
    method: mean
  - name: DMNBuckner_5mm_mean
    kind: SphereAggregation
    coords: DMNBuckner
    radius: 5
    method: mean

Storage#

Finally, we need to define how and where the results will be stored. This is done using the storage section, which must be configured using the kind key to specify the storage to use, and additional keys to pass parameters.

For example, to use the HDF5FeatureStorage storage, we just need to specify where we want to store the results:

storage:
  kind: HDF5FeatureStorage
  uri: /data/junifer/example/oasis_vbm_testing.hdf5

5.1.3. Complete Example#

This is how the full Oasis VBM Testing dataset example configuration file looks like:

with: junifer.testing.registry

workdir: /tmp

datagrabber:
  kind: OasisVBMTestingDataGrabber

markers:
  - name: Schaefer100x7_mean
    kind: ParcelAggregation
    parcellation: Schaefer100x7
    method: mean
  - name: Schaefer200x7_mean
    kind: ParcelAggregation
    parcellation: Schaefer200x7
    method: mean
  - name: DMNBuckner_5mm_mean
    kind: SphereAggregation
    coords: DMNBuckner
    radius: 5
    method: mean

storage:
  kind: HDF5FeatureStorage
  uri: /data/junifer/example/oasis_vbm_testing.hdf5