This is an old revision of the document!

How can I share my MEG data?

Sharing data along with published results is a vital step towards better reproducibility of MEG research and supports the ongoing development and validation of new analysis methods.

Funding agencies and journals are requiring more and more that data is being shared.

You may have an institutional data repository or a national data repository. Alternatively, you can consider using Zenodo, Harvard DataVerse, or OpenFMRI, which all accept submissions that suit the requirements of raw and minimally processed MEG data.

Most data repositories will create a persistent identifier linked to your dataset, such as a DOI or Handle. The dataset can be cited using the title, list of authors and the persistent handle.

Even better than only sharing the data is to write a accompanying publication that describes the dataset in detail and upload the data with the publication. There are dedicated peer-reviewed journals for data publications, among others Scientific Data (Nature), GigaScience. See here and here for some lists of data journals.

There is not a simple answer to this, but ideally you would share:

  • raw MEG data in the original vendor-specific format
  • defaced anatomical MRI with the coregistration information between MRI and MEG
  • if applicable the mapping of trigger codes on experimental conditions (stimuli and responses) and presentation log files
  • demographics (age, sex)

Furthermore, you can consider sharing the following:

  • minimally processed data
  • specification of bad channels and segments
  • cortical sheet source models (e.g. obtained from FreeSurfer)
  • volume conduction models (e.g. the boundaries that define brain, skull and scalp)

Here it helps to clarify some often used terminology:

  • The identity of the subject relates to personally identifiable information that is stored in, or linked to the (biological, structural and behavioral) research data.
  • Anonymous means that it is impossible to link the data to the identity of the subject in any way.
  • Pseudonomized means that the link between data and subject identity is provided as a (symbolic) identifier, where the key that links the pseudonym to actual identity is only known to the original researcher that acquired the data.
  • Deidentified means that identifying features have been removed.

The MEG and corresponding imaging data should be pseudonomized (i.e. using subject codes instead of names) and deidentified (i.e. no personal identifiable information contained in the files or data). For both MEG and MRI that means that the subject name and exact date of birth should not be stored in the header of the MEG dataset. For the MRI, the identifiable features of the imaging data (i.e. the face) should be removed.

There is not per se a perfect format for sharing the dataset, so you have to be pragmatic and where needed consider the peculiarities of your dataset. We recommend that you follow the Brain Imaging Data Structure (BIDS), for which an introduction has been published here.

The MEG extension is now part of the main BIDS specification, see here. Other extensions are still in draft status, but sufficiently well-defined to provide guidance. See this Google doc for EEG and this one for intracranial EEG, including sEEG and ECoG.