BayesianGroupAna.py


usage: /home/afniHQ/afni.build/pub.dist/bin/linux_ubuntu_16_64/BayesianGroupAna.py

[-h] -dataTable DATATABLE -y VAR [-help] [-prefix PREFIX] [-x VAR [VAR …]] [-no_center] [-iterations ITER] [-chains CHAINS] [-control_list LIST] [-plot] [-more_plots TYPE [TYPE …]] [-RData] [-seed SEED] [-overwrite]

Overview

This program conducts Bayesian Group Analysis (BGA) on a list
(e.g., 6 or more) of regions of interest (ROIs) as laid out in Chen et al.
(2018, https://www.biorxiv.org/content/early/2018/02/20/238998).
Compared to the conventional univariate GLM in which each voxel or ROI is
considered autonomous and analyzed independently, BGA pools and shares the
information across the ROIs in a multilevel system. It is the
probability of incorrect sign, instead of false positive rate that is
controlled. In other words, there is only one BGA model that incorporates
the data from all ROIs.

This will explore the effect of X on Y at each ROI. The computation may
take a few minutes or more depending on the amount of input data and
model complexity. The final inferences are conducted through the
posterior distribution or quantile intervals for each effect that are
provided in a table in the output. A boxplot can also be generated if
requested with -plot.

The computation requires that the R package "brms" be installed
(e.g., through rPkgsInstall).
More info on the brms package can be found here:
    https://CRAN.R-project.org/package=brms
And the brms reference manual is here:
    https://cran.r-project.org/web/packages/brms/brms.pdf

Details

Similar to 3dMVM and 3dLME, a data table  should be created containing
the input data and relevant variables (with at least 3 columns: subject
labels, ROI labels, response variable values).

The -dataTable should be formatted as follows:

    Subj  ROI   some_y  some_x other_x
    S001  roi1  0.12    0.056  0.356
    S001  roi2  0.65    0.232  0.231
    S002  roi1  0.14    0.456  0.856
    S002  roi2  0.64    0.432  0.431

The Subj and ROI columns must be included with the exact spelling!!
If there are no x variables, only the intercept will be calculated.

Outputs

Given -prefix is "gangBGA" and -x is "some_x", the default outputs are the
following files:

    gangBGA_summary.txt:
        Summary of the brmsfit object from R.

    gangBGA_rhats.csv:
        rhats for each effect and x variable combination.

    gangBGA_Intercept_table.csv:
        Table with the MedianEst, StdDev, 2.50%, 5%, 50%, 95%, and 97.50%
        of each ROI for the Intercept term.

    gangBGA_some_x_table.csv:
        The same table as the Intercept but for the some_x variable.

Caveats

All x variables are centered by default.

The boxplot with -plot is not a standard boxplot.
It is a plot of the 2.50%, 5%, 50%, 95%, 97.50% percentiles.
The coloring of the boxes is determined by where the zero line crosses the
box and whiskers.
    White:  The zero line crosses the main box (between 5% and 95%).
    Purple: The zero line crosses between the whiskers and the main box.
            (2.50% to 5%) OR (95% to 97.50%)
    Red:    The zero line does not cross the box or the whiskers.

Additional plot types for -more_plots include (not sure all of these work):
    hist dens hist_by_chain dens_overlay violin intervalsareas
    acf acf_bar trace trace_highlight rhat rhat_hist neff neff_hist
    nuts_acceptance nuts_divergence nuts_stepsize nuts_treedepth
    nuts_energy

Tables and plots will be created for the intercept and all specified x
variables separately. So there may be a lot of output.

Examples

Minimum requirement only calculates the intercept (may not be useful).
    BayesianGroupAna.py -dataTable my_roi_data.txt -y zscore

More useful. Calculates 2 x variables and saves out some plots.
    BayesianGroupAna.py -dataTable my_roi_data.txt  \
                        -prefix dock_of_the_bayes   \
                        -y zscore -x some_x other_x \
                        -chains 4 -iterations 1000  \
                        -plot -more_plots rhat violin

Options

Required arguments:
  -dataTable DATATABLE  Input text file.
  -y VAR                Column name for the y variable.

Optional arguments:
  -h, --help            show this help message and exit
  -help                 Show this help.
  -prefix PREFIX        Name of the output file.
  -x VAR [VAR ...]      Column name for the x variables. If not specified,
                        only the intercept will be added.
  -no_center            Disable centering on the x variables. Maybe useful if
                        you centered manually.
  -iterations ITER      Number of total iterations per chain including warmup.
                        Default [1000]
  -chains CHAINS        Number of Markov chains. Default [4]
  -control_list LIST    Comma separated list of control parameters to pass to
                        the brm function. (example:
                        'adapt_delta=0.99,max_treedepth=20'). Default is the
                        brm function defaults
  -plot                 Output box, fit, and posterior prediction plots.
  -more_plots TYPE [TYPE ...]
                        Output "stanplots" given different types of plot
                        names.
  -RData                Save the R session workspace and data.
  -seed SEED            Seed to generate random number. Default [1234]
  -overwrite            Overwrites the output files.

Justin Rajendra circa 05/2018
4 Gang Box...
Keep on keeping on!