proposeNCP: Number of independent components proposal function
In mlica2: Independent Component Analysis using Maximum Likelihood

Description Usage Arguments Value Author(s) References Examples

View source: R/proposeNCP.R

This function takes the output of 'PriorNormPCA' and returns for a given threshold the number of components to be inferred for subsequent ICA.

1	proposeNCP(prPCA, thresh = 0.1)

`prPCA`	The output object from 'PriorNormPCA'.
`thresh`	Threshold on eigenvalues.

A list with following components:

X: Normalised data matrix.

x: Normalised data matrix projected onto selected subspace.

pEx: Selected eigenvectors defining subspace for projection.

pCorr: Projected correlation matrix.

ncp: The dimension of the selected subspace(=number of independent components to be inferred with subsequent ICA).

Andrew Teschendorff a.teschendorff@ucl.ac.uk

Hyvaerinen A., Karhunen J., and Oja E.: Independent Component Analysis, John Wiley and Sons, New York, (2001).

Kreil D. and MacKay D. (2003): Reproducibility Assessment of Independent Component Analysis of Expression Ratios from DNA microarrays, Comparative and Functional Genomics *4* (3),300-317.

Liebermeister W. (2002): Linear Modes of gene expression determined by independent component analysis, Bioinformatics *18*, no.1, 51-60.

## The function is currently defined as
function (prPCA, thresh = 0.1) 
{
    X <- prPCA$X
    eigenvals.v <- diag(prPCA$Dx)
    Ex <- prPCA$Ex
    ntp <- nrow(X)
    ndim <- ncol(X)
    print("About to find ncp")
    p.cpts <- eigenvals.v[eigenvals.v > thresh]
    ncp <- length(p.cpts)
    pCorr <- diag(eigenvals.v[1:ncp])
    pEx <- Ex[, 1:ncp]
    x <- matrix(nrow = ntp, ncol = ncp)
    for (g in 1:ntp) {
        for (c in 1:ncp) {
            x[g, c] <- sum(X[g, ] * Ex[, c])/sqrt(diag(pCorr)[c])
        }
    }
    return(list(X = X, x = x, pEx = pEx, pCorr = pCorr, ncp = ncp))
  }