mixOmics (version 5.1.2)

tau.estimate: Optimal shrinkage intensity parameters.

Description

Estimation of the optimal shrinkage parameters as described in [1,2] and implemented in RGCCA [2]. The shrinkage estimates can be used in the rcc function instead of the tune.rcc which can be long and costly to compute with very large data sets. Note that both functions tune.rcc (which uses cross-validation) and tau.estimate (which uses the formula from [1]) may output different results. Note: we advise to center and scale the input matrices in rcc since the intput data for tau.estimate are centered and scaled to estimate the shrinkage parameters (as in [1]). The shrinkage parameters are also estimated using that function when running wrapper.rgcca when tau = 'optimal'. This esxplains the name of this function (tau.estimate), which may not fit with our current nomenclature with rcc.

Usage

tau.estimate(x)

Arguments

x

A list of matched data matrices on which the covariance matrix will be estimated.

Value

• tau
  Optimal shrinkage intensity parameter

References

[1] Schaefer J. and Strimmer K., 2005. A shrinkage approach to large-scale covariance matrix estimation and implications for functional genomics. Statist. Appl. Genet. Mol. Biol. 4:32.

[2] Jelizarow M., Guillemot V., Tenenhaus A., Strimmer K., Boulesteix A.-L., 2010. Over-optimism in bioinformatics: an illustration. Bioinformatics 26:1990-1998. [3] Tenenhaus A. and Tenenhaus M., (2011), Regularized Generalized Canonical Correlation Analysis, Psychometrika, Vol. 76(2), pp 257-284.

See Also

tune.rcc, rcc, wrapper.rgcca

Examples

## tuning the regularization parameters for 'rcc'
# ----------------------------------------------------
data(nutrimouse)
X <- nutrimouse$lipid
Y <- nutrimouse$gene

# create a list as an input
A = list(X, Y)

param = sapply(A, tau.estimate)
# regularisation parameters for X (first) and then Y
param

# note: we advise scaling the data as the rcc input in that case
rcc.result = rcc(scale(X), scale(Y), ncomp = 2, lambda1 = param[1], lambda2 = param[2])

# we observe quite a good deparation of the diets
plotIndiv(rcc.result, ind.names = nutrimouse$diet, group = nutrimouse$genotype)