In this repository, we use two methods to correct the upward bias in the naive calculation of information content, seeking a bias-free estimation of the information carried by neuronal activity about a given encoded variable.
The repository includes two methods for correcting the bias in the naive calculation of information content for limited sample sizes:
1. The scaled shuffle reduction (SSR) method - estimating the ratio between the bias in the naive and shuffle information and subtracting from the naïve information the shuffle information scaled by the estimated bias ratio.
2. The bounded asymptotic extrapolation (BAE) method - fitting the function of how the naive information changes with sample size and extrapolating it to infinity.
The naive estimation of information content is based on either the Skaggs information index (SI) or on the mutual information (MI).
Scripts are provided in the Scripts directory. Sample data sets for both real and simulated data are provided in the sample data directory.
The full data sets for both real and simulated data are provided in the Full data set directory.
To perform the analysis on the real or simulated sample data, call the estimate_unbiased_information.m function. The settings for calling this function are defined in the call_function_estimate_unbiased_information.m script.
An alternative option it to use the demo_real_data.m or demo_simulated_data.m scripts.
Before running this script, change the data_pathway to the sample data directory on your computer.
-
spike_train - a matrix of size TxN, where T is the number of time bins and N is the number of neurons. Each element is the spike count of a given neuron in a given time bin.
-
stimulus_trace - a vector of size T, where each element is the stimulus value in a given time bin.
-
General parameters (dt, number of shuffles, subsampling repetitions, etc.)
-
Data statistics (average rates, average active time bins, fraction of significant cells, etc.)
-
Estimated information (SI and MI based on the SSR and BE methods)
The bias correction methods can be modified to accomodate additional information theoretic measures.
The obtained results of the SSR and BAE methods are compared against shuffle reduction (SR) and asymptotic extrapolation (AE).
When applied to quantify the spatial tuning of place cells, the common practice is to first filter out time bins durning which the animal is not running (i.e., analyze only time bins with speed > threshold)