LsOPLS: Light-sparse-OPLS method for biomarker discovery
In ManonMartin/MBXUCL: R Routines for metabolomic data

Description Usage Arguments Value Examples

Implementation of the Light-sparse-OPLS algorithm Application of the sPLS method to a deflated matrix obtained by OPLS (matrix without the Y-orthogonal components) Cross-validated two-step optimization for the numbers of orthogonal and predictive components.

1	LsOPLS(X, Y, Auto, No, cv, Np, Eta, Method)

`X`	A numeric matrix containing the spectra.
`Y`	A numeric vector containing the target to predict.
`Auto`	Argument (TRUE or FALSE). If TRUE, the first optimalized step of L-sOPLS is performed. If FALSE, No has to be chosen.
`No`	The number (integer) of orthogonal components to extract (with a maximum of 9 orthogonal components). If Auto=TRUE, No must be such that No=NA.
`cv`	The number of cross-validation segments (for LOO: cv = dim(X)[1] )
`Np`	The number (integer) of predictive components in the L-sOPLS. Can be a sequence of values to test in the optimization step (ex: seq(1,5,by=1)).
`Eta`	Parameter for the level of sparsity into the sPLS step. Can be a sequence of values to test in the optimization step (ex: seq(0.6,0.99,by=0.01)).
`Method`	Parameter for the choice of the PLS algorithm ("pls2" or "simpls")

A print of cross-validated optimal parameters, a grid of RMSEP values and the sPLS final step containing the list of final selected biomarkers:

The number of final orthogonal dimensions (No)

Cross-validation results (containing a MSPE matrix, the optimal Eta parameter $eta.opt and the optimal number of predictive dimensions $K.opt)

A RMSEP matrix

Sparse results with the final selected variables

data("HumanSerum")
X <- as.matrix(HumanSerumSpectra)
Ex1 <- LsOPLS(X, ClassHS, Auto = TRUE, No = NA, cv = dim(X)[1],
Np = seq(1,3,by=1), Eta = seq(0.8,0.99,by=0.01), Method = "simpls")
Ex2 <- LsOPLS(X, ClassHS, Auto = FALSE, No = 2, cv = dim(X)[1],
Np = seq(1,3,by=1), Eta = seq(0.8,0.99,by=0.01), Method = "pls2")
print(Ex1)
print(Ex2)