Supervised Dimensional Reduction by Guided Partial Least Squares

Description

Guided partial least squares (guided-PLS) is the combination of partial least squares by singular value decomposition (PLS-SVD) and guided principal component analysis (guided-PCA). For the details of the methods, see the reference section of GitHub README.md <https://github.com/rikenbit/guidedPLS>.

Details

The DESCRIPTION file:

Index of help topics:

dummyMatrix             Toy model data for using dNMF, dSVD, dsiNMF,
                        djNMF, dPLS, dNTF, and dNTD
guidedPLS               Guided Partial Least Squares (guied-PLS)
guidedPLS-package       Supervised Dimensional Reduction by Guided
                        Partial Least Squares
PLSSVD                  Partial Least Squares by Singular Value
                        Decomposition (PLS-SVD)
softThr                 Soft-thresholding to make a sparse vector
                        sparse
sPLSDA                  Sparse Partial Least Squares Discriminant
                        Analysis (sPLS-DA)
toyModel                Toy model data for using PLSSVD, sPLSDA, and
                        guidedPLS

Author(s)

NA

Maintainer: NA

References

Le Cao, et al. (2008). A Sparse PLS for Variable Selection when Integrating Omics Data. Statistical Applications in Genetics and Molecular Biology, 7(1)

Reese S E, et al. (2013). A new statistic for identifying batch effects in high-throughput genomic data that uses guided principal component analysis. Bioinformatics, 29(22), 2877-2883

See Also

toyModel,PLSSVD,sPLSDA,guidedPLS

Examples

ls("package:guidedPLS")

Partial Least Squares by Singular Value Decomposition (PLS-SVD)

Description

Two matrices X and Y sharing a row are required

Usage

PLSSVD(X, Y, k=.minDim(X, Y), cortest=FALSE,
  deflation=FALSE, fullrank=TRUE, verbose=FALSE)

Arguments

Value

scoreX : Score matrix which has M-rows and K-columns. loadingX : Loading matrix which has N-rows and K-columns. scoreY : Score matrix which has L-rows and K-columns. loadingY : Loading matrix which has N-rows and K-columns. d : K-length singular value vector of the cross-product matrix X'Y. corX: Correlation Coefficient (Default: NULL) corY: Correlation Coefficient (Default: NULL) pvalX: P-value vector of corX (Default: NULL) pvalY: P-value vector of corY (Default: NULL) qvalX: Q-value vector of BH method against pvalX (Default: NULL) qvalY: Q-value vector of BH method against pvalY (Default: NULL)

Author(s)

Koki Tsuyuzaki

References

Le Cao, et al. (2008). A Sparse PLS for Variable Selection when Integrating Omics Data. Statistical Applications in Genetics and Molecular Biology, 7(1)

Examples

# Test data
data <- toyModel()

# Simple usage
out <- PLSSVD(X=data$X1, Y=data$Y1, k=4)

Toy model data for using dNMF, dSVD, dsiNMF, djNMF, dPLS, dNTF, and dNTD

Description

A label vector is converted to a dummy matrix.

Usage

dummyMatrix(y, center=TRUE)

Arguments

Value

A matrix is generated. The number of row is equal to the length of y and the number of columns is the number of unique elements of y.

Author(s)

Koki Tsuyuzaki

Examples

y <- c(1, 3, 2, 1, 4, 2)
dummyMatrix(y)

Guided Partial Least Squares (guied-PLS)

Description

Four matrices X1, X2, Y1, and Y2 are required. X1 and Y1 are supposed to share the rows, X2 and Y2 are supposed to share the rows, and Y1 and Y2 are supposed to share the columns.

Usage

guidedPLS(X1, X2, Y1, Y2, k=.minDim(X1, X2, Y1, Y2),
    cortest=FALSE, fullrank=TRUE, verbose=FALSE)

Arguments

Value

res: object of svd() loadingYX1: Loading vector to project X1 to lower dimension via Y1 (M times k). loadingYX2: Loading vector to project X2 to lower dimension via Y2 (P times k). scoreX1: Projected X1 (N times k) scoreX2: Projected X2 (O times k) scoreYX1: Projected YX1 (L times k) scoreYX2: Projected YX2 (L times k) corYX1: Correlation Coefficient (Default: NULL) corYX2: Correlation Coefficient (Default: NULL) pvalYX1: P-value vector of corYX1 (Default: NULL) pvalYX2: P-value vector of corYX2 (Default: NULL) qvalYX1: Q-value vector of BH method against pvalYX1 (Default: NULL) qvalYX2: Q-value vector of BH method against pvalYX2 (Default: NULL)

Author(s)

Koki Tsuyuzaki

References

Le Cao, et al. (2008). A Sparse PLS for Variable Selection when Integrating Omics Data. Statistical Applications in Genetics and Molecular Biology, 7(1)

Reese S E, et al. (2013). A new statistic for identifying batch effects in high-throughput genomic data that uses guided principal component analysis. Bioinformatics, 29(22), 2877-2883

Examples

# Test data
data <- toyModel()

# Simple usage
out <- guidedPLS(X1=data$X1, X2=data$X2, Y1=data$Y1, Y2=data$Y2, k=4)

Sparse Partial Least Squares Discriminant Analysis (sPLS-DA)

Description

Two matrices X and Y sharing a row are required

Usage

sPLSDA(X, Y, k=.minDim(X, Y), cortest=FALSE, lambda=1, thr=1e-10, fullrank=TRUE,
    num.iter=10, verbose=FALSE)

Arguments

Value

scoreX : Score matrix which has M-rows and K-columns. loadingX : Loading matrix which has N-rows and K-columns. scoreY : Score matrix which has L-rows and K-columns. loadingY : Loading matrix which has N-rows and K-columns. d : K-length singular value vector of the cross-product matrix X'Y. corX: Correlation Coefficient (Default: NULL) corY: Correlation Coefficient (Default: NULL) pvalX: P-value vector of corX (Default: NULL) pvalY: P-value vector of corY (Default: NULL) qvalX: Q-value vector of BH method against pvalX (Default: NULL) qvalY: Q-value vector of BH method against pvalY (Default: NULL)

Author(s)

Koki Tsuyuzaki

References

Le Cao, et al. (2008). A Sparse PLS for Variable Selection when Integrating Omics Data. Statistical Applications in Genetics and Molecular Biology, 7(1)

Examples

# Test data
data <- toyModel()

# Simple usage
out <- sPLSDA(X=data$X1, Y=data$Y1, k=4)

Soft-thresholding to make a sparse vector sparse

Description

The degree of the sparseness of vector is controlled by the lambda parameter.

Usage

softThr(y, lambda=1)

Arguments

Value

A numerical vector, whose length is the same as that of y.

Author(s)

Koki Tsuyuzaki

Examples

y <- seq(-2, 2, 0.1)
softThr(y)

Toy model data for using PLSSVD, sPLSDA, and guidedPLS

Description

The data is used for confirming the algorithm are properly working.

Usage

toyModel(model="Easy", seeds=123)

Arguments

Value

A list object containing a set of matrices X1, X2, Y1, Y1_dummy, Y2, Y1_dummy.

Author(s)

Koki Tsuyuzaki

See Also

PLSSVD,sPLSDA,guidedPLS

Examples

data <- toyModel(seeds=123)