cpm.lesion

CPM with the leave-one-network-out lesion approach

cpm.lesion(train.data, test.data, train.outcome, test.outcome, p = 0.05)

Arguments

train.data

An N x E matrix containing the vectorized edges in the training dataset; where N = number of subjects, E=number of edges

test.data

An N x E matrix containing the vectorized edges in the training dataset

train.outcome

The outcome variable to predict, within the training dataset

test.outcome

The outcome variable to predict, within the testing dataset

p

The p-value threshold of a Pearson's correlation test between the feature and outcome that determines if the feature is selected. Set to 0.05 by default.

Value

Returns a matrix with the following columns

• lesion.model The CPM model in which the listed network is left out

• positive The predicted-actual correlations for each of the lesion.models' positive network model

• negative The predicted-actual correlations for each of the lesion.models' negative network model

• both The predicted-actual correlations for each of the lesion.models' combined positive + negative network model

Details

This function runs the cpm.train() and cpm.predict() while leaving out a network of edges each time. The changes in predicted-actual correlation while a network being left out can be interpreted to be an indication of the contribution of the network to predicting the outcome

Examples

if (FALSE) {
model1.lesion=cpm.lesion(train.data=FC_data.train,test.data=FC_data.test,train.outcome=train_dat$age, test.outcome=test_dat$age,p=0.05)
}