drCCAcombine: A function to combine several data sets

Description Usage Arguments Details Value Author(s) References See Also Examples

Description

Performs drCCA on a collection of data sets with co-occurring samples. The method utilizes regularized canonical correlation analysis to find linear projections for each of the data sets, and uses those to construct a combined representation of lower dimensionality than the original collection. The method suggests a specific dimensionality for the combined representation, but it is possible to obtain also combined data sets of different dimensionality.

Usage

1
drCCAcombine(datasets, reg=0, nfold=3, nrand=50)

Arguments

datasets

A list containing the data matrices to be combined. Each matrix needs to have the same number of rows (samples), but the number of columns (features) can differ. Each row needs to correspond to the same sample in every matrix.

reg

Regularization parameter for the whitening step used to remove data-set specific variation. The value of parameter must be between 0 and 1. The default value is set to 0, which means no regularization will be used. If a non-zero value is given it means that some of the dimensions with the lowest variance are ignored when whitening. In more technical terms, the dimensions whose total contribution to the sum of eigenvalues of the covariance matrix of each data set is below reg will not be used for the whitening.

nfold

The number of cross-validation folds used in the automatic dimensionality estimation process. The default value is 3.

nrand

The number of random comparison data-sets created for the automatic dimensionality estimation process. The default value is 50.

Details

The function uses regCCA to perform the canonical correlation analysis. The dimensionality of the combined data set is selected using a statistical test that aims to find which dimensions capture shared variation significantly more than what would be found under the assumption that the data sets were independent. For this purpose rnand collections of random matrices with similar variance structure but no between-data dependencies are created. The amount of variation each dimension extracts from leave-out data in the cross-validation setting with nfold folds is compared to the distribution obtained from the random matrices, and the dimensions that differ significantly from the null hypothesis of independence are kept in the combined representation. For details, please check the reference.

Value

The function returns a list of two values.

proj

The representation obtained by combining the source data sets. This is a matrix that contains a feature representation for each of the samples in the analyzed collection. Each row in this result matches the corresponding row in the original data sets.

n

The number of dimensions in the combined representation. This is equal to ncol(proj).

Author(s)

Abhishek Tripathi abhishektripathi.at@gmail.com, Arto Klami

References

Tripathi A., Klami A., Kaski S. (2007), Simple integrative preprocessing preserves what is shared in data sources.

See Also

regCCA

Examples

1
2
3
    # data(expdata1)
    # data(expdata2)
    # drCCAcombine(list(expdata1,expdata2),0,2,3)

dmt documentation built on May 2, 2019, 5:49 p.m.

Related to drCCAcombine in dmt...