This description refers to the packages available at 

                 http://www.lcs.poli.usp.br/~baccala/BIHExtension2014

This set of Matlab packages and ad hoc routines were used to simulate three 
toy-models and generate figures for the following manuscript submitted to 
Brain Informatics:

   K. Sameshima, D.Y. Takahashi and L.A. Baccala. On the statistical 
   performance of Granger-causal connectivity estimators. Brain 
   Informatics:Brain Data Computing and Health Studies, 2015. Submitted.

This is a three-part package containing Matlab codes and the specific simulation 
results (mat files) and analysis results, which were used to generate the 
figures for the manuscript:

1. AsympPDC Package version 3 contains routines and function for all three 
   forms of PDC (PDC, gPDC, iPDC) and DTF (DTF, DC, iDTF) estimations, as 
   well as their asymptotic statistics implementation. Future releases and 
   updates will be made available at

                 http://www.lcs.poli.usp.br/~baccala/pdc.

2. Our site contains the unmodified copy of MVGC v1, last updated on March 
   27 2014 13:50:46, and retrieved from 

                 http://www.sussex.ac.uk/sackler/mvgc/

   on December 09 2014, as present in 

      L. Barnett and A.K. Seth. The MCGV multivariate Granger causality 
      toolbox: A new approach to Granger-causal inference. J. Neurosci. 
      Methods, 223:50--68, 2014. 

                 http://dx.doi.org/10.1016/j.jneumeth.2013.10.018

3. Three toy-models simulation and analysis codes and data identified as:
   3.1. M1 - Model 1 stands for Closed-loop model, which corresponds to 
            Example 7 in Baccala and Sameshima. Overcoming the limitations 
            of correlation analysis for many simultaneously processed neural 
            structures, Progress in Brain Research,130:33--47, 2001.

                 http://dx.doi.org/10.1016/S0079-6123(01)30004-3

   3.2. M2 - Model 2 is a VAR(3) model corresponding to the Example 3 
            (p. 468) from Baccala & Sameshima. Partial directed coherence: 
            a new concept in neural structure determination. Biol. Cybern. 
            84:463-474, 2001.

                 http://dx.doi.org/10.1007/PL00007990

   3.3. M3 - Model 3 is a five-var VAR[3] model with large common exogenous 
            input, modified from Model 2, proposed by 
            Guo, Wu, Ding & Feng. Uncovering interactions in frequency 
            domains. PLoS Computational Biology, 4(5):1-10, February 8, 
            2008.

                 http://dx.plos.org/10.1371/journal.pcbi.1000087

            Note that the PDC results provided in this article are 
            incorrect. 

==========================================================================

- Set the Matlab path for the packages, i.e. set path for AsympPDC v3 and 
  MVGCv1 packages, and toy_models directory and its subdirectories.

- Make sure your Matlab installation has (1) Signal Processing, (2) Control
  System and (3) Statistics Matlab Toolboxes.  

- To perform new 1000 Monte Carlo simulations with the toy model, run
  the corresponding batch for corresponding model, i.e. either 
  m1_batch_simulations.m, m2_batch_simulations.m or m3_batch_simulations.m.
  Be aware that depending on the your particular computer power and the 
  model being simulated, it may take from several days to a week.

- After finishing the Monte Carlo simulation, please perform corresponding 
  analysis by running either m1_analysis.m, m2_analysis.m or m3_analysis.m.
  The analysis routine will generate an aggregated result mat-file, from a 
  series of files saved in m#_batch_simulations.m (where # = {1, 2 or 3}) for 
  different simulation data length, named 

       m1_monte_carlo_simulation_results_total.mat, 
       m2_monte_carlo_simulation_results_total.mat or 
       m3_monte_carlo_simulation_results_total.mat, respectively. 

  Once the these files are generated, the subsequent analysis will use 
  these mat-files. The analysis routine will generate -log10(p-values) 
  boxplots for all measures (i.e. GCT, cMVGC, iPDC and iDTF) and  all 
  combination of scatterplots between -log10(p-values) of GCT, cMVGC, iPDC 
  and iDTF measures in function of simulation data length (i.e., 
  {100,200,500,1000,2000,5000,10000}). Note that only a subset of figures
  was used in the manuscript.
  
- In order to get the figures in the manuscript, please use the provided 
  results mat-file data, i.e.

       m1_monte_carlo_simulation_results_total.mat, 
       m2_monte_carlo_simulation_results_total.mat or 
       m3_monte_carlo_simulation_results_total.mat 

  to run the analysis routine. Actually, if these files are present in the 
  directory, the analysis routine will automatically pick one of these files.