In emeyers/NDTr: The Neural Decoding Toolbox in R

The NDTr uses two closely related data formats called raster format and binned format. For almost all analysis, one starts by saving data from each site in raster format. One then converts it to binned format using the create_binned_data() which has the data from all the sites at a coarser temporal resolution. The binned format data is then used in all decoding analyses. More information about what is required to have data in these specified formats is described below.

Raster format

Raster format data contains the data at the highest temporal resolution. For raster data, there is a separate file that contains a data frame of data each site (e.g., there is a separate file for each single neuron, each LFP channel, etc.). The reason for having data from each site in a separate file is to prevent memory one from running out of memory when trying to load data from many sites when the data is at a high temporal resolution.

For raster format data, the number of rows in the data frame correspond to the number of trials in the experiment. Data that is in raster format is a data frame that must contain variables (columns) that start with the following prefixes:

1. labels.XXX These variables contain labels of which experimental conditions were shown on a given trial.

2. time.XXX These variables contain the data for a given time.

There can also be two additional optional variables in a raster format data frame which are:

1. site_info.XXX These variables contain additional meta data out the site. For example, one could have a variable called site_info.brain_area which indicated which brain region a given site came from. All rows for a given site_info.XXX variables typically have the same value.

2. trial_number This is a variable that specifies a unique for each row indicating which trial a given row of data came from. This is useful for data where all sites were recorded simultaneously in order to allow one to do the decoding on actual simultaneously recorded data (e.g., by using the ds_basic() create_simultaneously_recorded_populations argument).

The class attribute for data in raster format should be set to attr(raster_data, "class") <- c("raster_data", "data.frame"). This will enable the plot() function to correctly plot data in raster format.

Checking if data is in valid raster format

To test whether data correctly conforms to the requirements of raster format, one can use the internal function NDTr:::test_valid_raster_data_format().

Example raster-format data

Below is an example of raster format data file from the Zhang-Desimone 7 object data set.

raster_dir_name <- file.path(system.file("extdata", package = "NDTr"), "Zhang_Desimone_7object_raster_data_rda")
full_file_name <- file.path(raster_dir_name, "bp1001spk_01A_raster_data.rda")

# test the file is in valid raster format
NDTr::test_valid_raster_format(full_file_name)

# load the data to see the variables in it
load(full_file_name)
head(raster_data[, 1:10])

Binned format

Binned format data contains data from multiple sites (e.g., data from many neurons, LFP channels, etc.). Data that is in binned format is very similar to data that is in raster format except that it contains information from multiple sites and often contains the information at a coarser resolution. For example, binned data would typically contain firing rates over some time interval sampled at a lower rate, as opposed to raster format data that would typically contain individual spikes sampled at a higher rate. Binned format data is typically created from raster format data using the function create_binned_data which converts a directory of raster format files into a binned-format file that is used in subsequent decoding analyses.

Binned format data must be in a data frame where the number of rows in the data frame correspond to the number of trial in all experimental recording sessions across all sites. The binned format data frame must also contain the variables that start with the following prefixes:

1. siteID.XXX A unique number indicating a site a given row of data corresponds to.

2. labels.XXX These variables contain labels of which experimental conditions occurred on a given trial.

3. time.X_Y These variables contain data in a time range from X to Y.

There can also be two additional optional variables in a binned format data frame which are:

1. site_info.XXX These variables contain additional meta data out the site. For example, one could have a variable called site_info.brain_area which indicated which brain region a given site came from. All rows for a given site_info.XXX variables typically have the same value.

2. trial_number This is a variable that specifies a unique for each row indicating which trial a given row of data came from. This is useful for data where all sites were recorded simultaneously in order to allow one to do the decoding on actual simultaneously recorded data (e.g., by using the ds_basic() create_simultaneously_recorded_populations argument).

Checking if data is in valid binned format

To test whether data correctly conforms to the requirements of binned format, one can use the internal function NDTr:::test_valid_binned_data_format().

Example binned-format data

Below is an example of binned format data file from the Zhang-Desimone 7 object data set.

binned_file_name <- system.file("extdata/ZD_150bins_50sampled.Rda", package="NDTr")


# test the file is in valid binned format using an internal function
NDTr:::test_valid_binned_format(binned_file_name)


# load the data to see the variables in it
load(binned_file_name)
head(binned_data[, 1:10])

emeyers/NDTr documentation built on Aug. 8, 2020, 3:41 p.m.