Specification of experiment stimuli for sharing electrophysiology data
Jeffrey L. Teeters (UC Berkeley), Wayne Christopher (UC Berkeley), Marvin Thielk (UC Berkeley), Friedrich T. Sommer (UC Berkeley)
To develop a generalizable strategy for describing stimuli, we are using the data sets contributed to CRCNS.org as test cases. Specifically, we are developing methods to specify stimuli along with neural data using HDF5 as a storage container. The HDF5 schema we developed uses a hierarchy to organize the data into groups that correspond to the four main data types. The event type data consists of an array of event times, and an additional array containing parameter values that specify what event was at each time.
For visual stimuli, the parameters must indicate which image frame has appeared at each time point in the experiment. For audio stimuli, the parameters must map time points to positions in a sound file. Some issues we are addressing are:
1. Image frames in contributed data sets are specified in many different ways (for example: jpeg files, 2-D matrices, a script that generates an image sequence). For effective data sharing, there needs to be a standardized representation. We are investigating converting everything into standard 3-D arrays (x, y and time axis) stored in HDF5.
2. If applicable, parameters describing a stimulus content (for example, orientation of a bar or frequency and length of a pure tone) must also be stored.
3. To allow the same stimuli to be referenced from different experiments, it is advantageous to store the actual stimulus files apart from the files containing neural data. The reference between data file and stimuli files must be explicit and unambiguous.
4. Repetitions of the same stimuli (start of a repeat of a stimulus sequence) should also be easy to detect in order to analyze average responses across trial repeats.
Our goal is to develop generalizable schemes to store electrophysiology data and stimuli in HDF5 files so that both are accessible for online browsing and also for automated tools of machine learning.