R: Partial least squares

PLS-methods {Cardinal}

R Documentation

Partial least squares

Description

Performs partial least squares (also called projection to latent structures or PLS) on an imaging dataset. This will also perform discriminant analysis (PLS-DA) if the response is a factor. Orthogonal partial least squares options (O-PLS and O-PLS-DA) are also available.

Usage

## S4 method for signature 'SparseImagingExperiment,ANY'
PLS(x, y, ncomp = 3, method = c("pls", "opls"),
        center = TRUE, scale = FALSE,
        iter.max = 100, ...)

## S4 method for signature 'SparseImagingExperiment,ANY'
OPLS(x, y, ncomp = 3, ...)

## S4 method for signature 'PLS2'
predict(object, newx, newy, ncomp, ...)

## S4 method for signature 'PLS2'
fitted(object, ...)

## S4 method for signature 'PLS2'
summary(object, ...)

## S4 method for signature 'SImageSet,matrix'
PLS(x, y, ncomp = 3,
    method = "nipals",
    center = TRUE,
    scale = FALSE,
    iter.max = 100, ...)

## S4 method for signature 'SImageSet,ANY'
PLS(x, y, ...)

## S4 method for signature 'SImageSet,matrix'
OPLS(x, y, ncomp = 3,
    method = "nipals",
    center = TRUE,
    scale = FALSE,
    keep.Xnew = TRUE,
    iter.max = 100, ...)

## S4 method for signature 'SImageSet,ANY'
OPLS(x, y, ...)

## S4 method for signature 'PLS'
predict(object, newx, newy, ...)

## S4 method for signature 'OPLS'
predict(object, newx, newy, keep.Xnew = TRUE, ...)

Arguments

`x`	The imaging dataset on which to perform partial least squares.
`y`	The response variable, which can be a `matrix` or a `vector` for ordinary PLS, or a `factor` or a `character` for PLS-DA.
`ncomp`	The number of PLS components to calculate.
`method`	The function used to calculate the projection.
`center`	Should the data be centered first? This is passed to `scale`.
`scale`	Shoud the data be scaled first? This is passed to `scale`.
`iter.max`	The number of iterations to perform for the NIPALS algorithm.
`...`	Passed to the next PLS method.
`object`	The result of a previous call to `PLS`.
`newx`	An imaging dataset for which to calculate their PLS projection and predict a response from an already-calculated `PLS` object.
`newy`	Optionally, a new response from which residuals should be calcualted.
`keep.Xnew`	Should the new data matrix be kept after filtering out the orthogonal variation?

Value

An object of class PLS2, which is a ImagingResult, or an object of class PLS, which is a ResultSet. Each elemnt of resultData slot contains at least the following components:

fitted:: The fitted response.
loadings:: A matrix with the explanatory variable loadings.
weights:: A matrix with the explanatory variable weights.
scores:: A matrix with the component scores for the explanatary variable.
Yscores:: A matrix objects with the component scores for the response variable.
Yweights:: A matrix objects with the response variable weights.
coefficients:: The matrix of the regression coefficients.

The following components may also be available for classes OPLS and OPLS2.

Oloadings:: A matrix objects with the orthogonal explanatory variable loadings.
Oweights:: A matrix with the orthgonal explanatory variable weights.

If y is a categorical variable, then a categorical class prediction will also be available in addition to the fitted numeric response.

Author(s)

Kylie A. Bemis

References

Trygg, J., & Wold, S. (2002). Orthogonal projections to latent structures (O-PLS). Journal of Chemometrics, 16(3), 119-128. doi:10.1002/cem.695

Examples

setCardinalBPPARAM(SerialParam())

set.seed(1)
x <- simulateImage(preset=2, npeaks=10, dim=c(10,10),
    snoise=1, sdpeaks=1, representation="centroid")

y <- makeFactor(circle=pData(x)$circle, square=pData(x)$square)

pls <- PLS(x, y, ncomp=1:3)

summary(pls)

opls <- OPLS(x, y, ncomp=1:3)

summary(pls)

[Package Cardinal version 2.6.0 Index]