cpm {NetworkToolbox}

Connectome-based Predictive Modeling

Description

Suite of functions for Connectome-based Predictive Modeling (CPM). See and cite Finn et al., 2015; Rosenberg et al., 2016; Shen et al., 2017

• cpmIV

  Internal Validation method (Rosenberg et al., 2016; Shen et al., 2017). Using a leave-one-out approach, this method correlates a behavioral statistic bstat with each edge of a whole-brain network across participants. Using the significant edges in the network thresh, a connectome model is built (without the participant's network). A linear regression model is fit, with the behavioral statistic being regressed on the connectome model. The left out participants connectome model is then used with the linear regression weights to compute their predicted behavioral score. This is repeated for every participant. The predicted scores are correlated with their observed score. Significant values suggest that the connectome is related to the behavioral statistic

• cpmIVperm

  Performs a permutation test of the results obtained by cpmIV. The permutation test quantifies whether the results obtained by the original cpmIV are significantly different than a random model (see Shen et al., 2017)

• cpmEV

  UNDER DEVELOPMENT. External Validation method (Beaty et al., 2018). Performs similar function as cpmIV but uses data to train train_na the connectome model using a behavioral statistic train_b. This training connectome model is then used to predict another dataset valid_na, using the same behavioral statistic valid_b. The full training dataset FALSE or the leave-one-out overlap = TRUE approach can be used

• cpmFP

  Fingerprinting method (Finn et al., 2015). Uses CPM approach to identify participants across two sessions

• cpmFPperm

  Fingerprinting method (Finn et al., 2015). Uses permutation method to estimate the significance of of the cpmFP results

• cpmPlot

  Plots the CPM results

Usage

cpmIV(neuralarray, bstat, kfolds, covar, thresh = .01,
      connections = c("separate", "overall"), groups = NULL,
      method = c("mean", "sum"), model = c("linear","quadratic","cubic"),
      corr = c("pearson","spearman"), nEdges, 
      standardize = FALSE, cores, progBar = TRUE, plots = TRUE)
      
cpmIVperm(iter = 1000, ...)
      
cpmEV(train_na, train_b, valid_na, valid_b, thresh = .01,
      overlap = FALSE, progBar = TRUE)
      
cpmFP(session1, session2, progBar = TRUE)

cpmFPperm(session1, session2, iter = 1000, progBar = TRUE)

cpmPlot(cpm.obj, visual.nets = FALSE)

Arguments

neuralarray

Array from convertConnBrainMat function

bstat

Behavioral statistic for each participant with neural data (a vector)

kfolds

Numeric. Number of k-fold validation samples. Defaults to the number of participants in the sample (i.e., n), which is also known as leave-one-out validation. Recommended folds are 5 and 10

covar

Covariates to be included in predicting relevant edges (time consuming). Must be input as a list() (see examples)

thresh

Sets an \alpha threshold for edge weights to be retained. Defaults to .01

connections

Character. Should positive and negative correlations be separated or used together? Defaults to "separate"

groups

Allows grouping variables to be used for plotting points. Must be a vector. Defaults to NULL

method

Use "mean" or "sum" of edge strengths in the positive and negative connectomes. Defaults to "mean"

model

Regression model to use for fitting the data. Defaults to "linear"

corr

Correlation method for assessing the relationship between the behavioral measure and edges between ROIs. Defaults to "pearson". Set to "spearman" for non-linear or monotonic associations

nEdges

Number of participants that are required to have an edge to appear in the plots. Defaults to 10 percent of edges in participants

standardize

Should the behavioral statistic (bstat) be standardized? Defaults to FALSE

cores

Number of computer processing cores to use when performing covariate analyses. Defaults to n - 1 total number of cores. Set to any number between 1 and maximum amount of cores on your computer

progBar

Should progress bar be displayed? Defaults to TRUE. Set to FALSE for no progress bar

plots

Should plots be plotted? Defaults to TRUE. Set to FALSE to hide plots

train_na

Training dataset (an array from convertConnBrainMat function)

train_b

Behavioral statistic for each participant for the training neural data (a vector)

valid_na

Validation dataset (an array from convertConnBrainMat function)

valid_b

Behavioral statistic for each participant for the validation neural data (a vector)

overlap

Should leave-one-out cross-validation be used? Defaults to FALSE (use full dataset, no leave-one-out). Set to TRUE to select edges that appear in every leave-one-out cross-validation network (time consuming)

session1

Array from convertConnBrainMat function (first session)

session2

Array from convertConnBrainMat function (second session)

iter

Number of iterations to perform. Defaults to 1000

cpm.obj

cpm object

visual.nets

Boolean. Uses qgraph to plot connectivity between the networks as a network. Defaults to FALSE. Set to TRUE to visualize the networks

...

Additional arguments to be passed from a cpm function

Value

cpmIV and cpmEV:

Returns a list containing:

results

A matrix containing: r coefficient (r), p-value (p-value), mean absolute error (mae), root mean square error (rmse)

posMask

Positive connectivity for input in BioImage Suite Connectivity Viewer

negMask

Negative connectivity for input in BioImage Suite Connectivity Viewer

cpmIVperm:

Returns a matrix containing p-values for positive and negative prediction models

cpmFP:

Returns a matrix containing the percentage and number of correctly identified subjects for sessions 1 and 2

cpmPlot:

Returns plot of connectivity differences between the positive and negative masks

Author(s)

Alexander Christensen <[e-mail]>

References

Beaty, R. E., Kenett, Y. N., Christensen, A. P., Rosenberg, M. D., Benedek, M., Chen, Q., Fink, A., Qiu, J., Kwapil, T. R., Kane, M. J., & Silvia, P. J. (2018). Robust prediction of individual creative ability from brain functional connectivity. Proceedings of the National Academy of Sciences, 115, 1087-1092.

Finn, E. S., Shen, X., Scheinost, D., Rosenberg, M. D., Huang, J., Chun, M. M., Papademetris, X., Constable, R. T. (2015). Functional connectome fingerprinting: Identifying individuals using patterns of brain connectivity. Nature Neuroscience, 18, 1664-1671.

Rosenberg, M. D., Finn, E. S., Scheinost, D., Papademetris, X., Shen, X., Constable, R. T., Chun, M. M. (2016). A neuromarker of sustained attention from whole-brain functional connectivity. Nature Neuroscience, 19, 165-171.

Shen, X. Finn, E. S., Scheinost, D., Rosenberg, M. D., Chun, M. M., Papademetris, X., Constable, R. T. (2017). Using connectome-based predictive modeling to predict individual behavior from brain connectivity. Nature Protocols, 12, 506-518.

Wei, T. & Simko, V.(2017). R package "corrplot": Visualization of a correlation matrix (Version 0.84).

Examples

# Load data
behav <- behavOpen

## Not run: 

# Create path to temporary file
temp <- tempfile()

# Download to temporary file
googledrive::drive_download(
paste("https://drive.google.com/file/d/",
"1T7_mComB6HPxJxZZwwsLLSYHXsOuvOBt",
"/view?usp=sharing", sep = ""),
path = temp
)

# Load resting state brain data
load(temp)

# Run cpmIV
res <- cpmIV(neuralarray = restOpen, bstat = behav, cores = 4)

# Plot cpmIV results
cpmPlot(res)


## End(Not run)