Residuals {brainGraph}

Linear model residuals in structural covariance networks

Description

get.resid runs linear models across brain regions listed in a data.table (e.g., cortical thickness), adjusting for variables in covars (e.g. age, sex, etc.), and calculates the externally Studentized (or leave-one-out) residuals.

The [ method reorders or subsets residuals based on a given numeric vector. However, this is used in bootstrap and permutation analysis and should generally not be called directly by the user.

The summary method prints the number of outliers per region, and the number of times a given subject was an outlier (i.e., across regions).

The plot method lets you check the model residuals for each brain region in a structural covariance analysis. It shows a qqplot of the studentized residuals, as output from get.resid.

Usage

get.resid(dt.vol, covars, method = c("comb.groups", "sep.groups"),
  use.mean = FALSE, exclude.cov = NULL, atlas = NULL, ...)

## S3 method for class 'brainGraph_resids'
x[i, g = NULL]

## S3 method for class 'brainGraph_resids'
summary(object, region = NULL,
  outlier.thresh = 2, ...)

## S3 method for class 'brainGraph_resids'
plot(x, region = NULL, outlier.thresh = 2,
  cols = FALSE, ids = TRUE, ...)

## S3 method for class 'brainGraph_resids'
nobs(object, ...)

## S3 method for class 'brainGraph_resids'
case.names(object, ...)

## S3 method for class 'brainGraph_resids'
groups(x)

## S3 method for class 'brainGraph_resids'
region.names(object)

## S3 method for class 'brainGraph_resids'
nregions(object)

Arguments

dt.vol

A data.table containing all the volumetric measure of interest (i.e., the object lhrh as output by import_scn)

covars

A data.table of the covariates of interest

method

Character string indicating whether to test models for subject groups separately or combined. Default: comb.groups

use.mean

Logical should we control for the mean hemispheric brain value (e.g. mean LH/RH cortical thickness). Default: FALSE

exclude.cov

Character vector of covariates to exclude. Default: NULL

atlas

Character string indicating the brain atlas

...

Arguments passed to brainGraph_GLM_design (optional)

x, object

A brainGraph_resids object

i

Numeric vector of the indices

g

Character string indicating the group. Default: NULL

region

Character vector of region(s) to focus on; default behavior is to show summary for all regions

outlier.thresh

Number indicating how many standard deviations above/below the mean indicate an outlier. Default: 2

cols

Logical indicating whether to color by group. Default: FALSE

ids

Logical indicating whether to plot subject ID's for outliers. Otherwise plots the integer index

Details

You can choose to run models for each of your subject groups separately or combined (the default) via the method argument. You may also choose whether to include the mean, per-hemisphere structural measure in the models. Finally, you can specify variables that are present in covars which you would like to exclude from the models. Optional arguments can be provided that get passed to brainGraph_GLM_design.

If you do not explicitly specify the atlas name, then it will be guessed from the size of your data. This could cause problems if you are using a custom atlas, with or without the same number of regions as a dataset in the package.

Value

get.resid - an object of class brainGraph_resids with elements:

data

A data.table with the input volume/thickness/etc. data as well as the covariates used in creating the design matrix.

X

The design matrix, if using default arguments. If use.mean=TRUE then it will be a named list with a separate matrix for the left and right hemispheres. If method='sep.groups', a nested named list for each group and hemisphere.

method

The input argument method

use.mean

The input argument use.mean

resids.all

The “wide” data.table of residuals

Group

Group names

atlas

The atlas name

summary.brainGraph_resids returns a list with two data tables, one of the residuals, and one of only the outlier regions

The plot method returns a trellis object or a list of ggplot objects

Note

It is assumed that dt.vol was created using import_scn. In older versions, there were issues when the Study ID was specified as an integer and was not “zero-padded”. This is done automatically by import_scn, so if you are using an external program, please be sure that the Study ID column is matched in both dt.vol and covars.

Author(s)

Christopher G. Watson, [e-mail]

See Also

influence.measures, qqnorm

Other Structural covariance network functions: Bootstrapping, IndividualContributions, brainGraph_permute, corr.matrix, import_scn, plot_volumetric

Examples

## Not run: 
myresids <- get.resids(lhrh, covars)
residPlots <- plot(myresids, cols=TRUE)

## Save as a multi-page PDF
ml <- marrangeGrob(residPlots, nrow=3, ncol=3)
ggsave('residuals.pdf', ml)

## End(Not run)