R/ipca.R

# ========================================================================================================
# The function ica.par and ica.def are borrowed from the fastICA package (see references).
# ========================================================================================================

#' Independent Principal Component Analysis
#' 
#' Performs independent principal component analysis on the given data matrix,
#' a combination of Principal Component Analysis and Independent Component
#' Analysis.
#' 
#' In PCA, the loading vectors indicate the importance of the variables in the
#' principal components. In large biological data sets, the loading vectors
#' should only assign large weights to important variables (genes, metabolites
#' ...). That means the distribution of any loading vector should be
#' super-Gaussian: most of the weights are very close to zero while only a few
#' have large (absolute) values.
#' 
#' However, due to the existence of noise, the distribution of any loading
#' vector is distorted and tends toward a Gaussian distribtion according to the
#' Central Limit Theroem. By maximizing the non-Gaussianity of the loading
#' vectors using FastICA, we obtain more noiseless loading vectors. We then
#' project the original data matrix on these noiseless loading vectors, to
#' obtain independent principal components, which should be also more noiseless
#' and be able to better cluster the samples according to the biological
#' treatment (note, IPCA is an unsupervised approach).
#' 
#' \bold{Algorithm} 1. The original data matrix is centered.
#' 
#' 2. PCA is used to reduce dimension and generate the loading vectors.
#' 
#' 3. ICA (FastICA) is implemented on the loading vectors to generate
#' independent loading vectors.
#' 
#' 4. The centered data matrix is projected on the independent loading vectors
#' to obtain the independent principal components.
#' @inheritParams pca
#' @param X a numeric matrix (or data frame).
#' @param ncomp integer, number of independent component to choose. Set by
#' default to 3.
#' @param mode character string. What type of algorithm to use when estimating
#' the unmixing matrix, choose one of \code{"deflation"}, \code{"parallel"}.
#' Default set to \code{deflation}.
#' @param fun the function used in approximation to neg-entropy in the FastICA
#' algorithm. Default set to \code{logcosh}, see details of FastICA.
#' @param max.iter integer, the maximum number of iterations.
#' @param tol a positive scalar giving the tolerance at which the un-mixing
#' matrix is considered to have converged, see fastICA package.
#' @param w.init initial un-mixing matrix (unlike fastICA, this matrix is fixed
#' here).
#' @return \code{ipca} returns a list with class \code{"ipca"} containing the
#' following components: \item{ncomp}{the number of independent principal
#' components used.} \item{unmixing}{the unmixing matrix of size (ncomp x
#' ncomp)} \item{mixing}{the mixing matrix of size (ncomp x ncomp)}
#' \item{X}{the centered data matrix} \item{x}{the independent principal
#' components} \item{loadings}{the independent loading vectors}
#' \item{kurtosis}{the kurtosis measure of the independent loading vectors}
#' \item{prop_expl_var}{Proportion of the explained variance of derived
#' components, after setting possible missing values to zero.}
#' @author Fangzhou Yao, Jeff Coquery, Kim-Anh Lê Cao, Florian Rohart, Al J Abadi
#' @seealso \code{\link{sipca}}, \code{\link{pca}}, \code{\link{plotIndiv}},
#' \code{\link{plotVar}}, and http://www.mixOmics.org for more details.
#' @references Yao, F., Coquery, J. and Lê Cao, K.-A. (2011) Principal
#' component analysis with independent loadings: a combination of PCA and ICA.
#' (in preparation)
#' 
#' A. Hyvarinen and E. Oja (2000) Independent Component Analysis: Algorithms
#' and Applications, \emph{Neural Networks}, \bold{13(4-5)}:411-430
#' 
#' J L Marchini, C Heaton and B D Ripley (2010). fastICA: FastICA Algorithms to
#' perform ICA and Projection Pursuit. R package version 1.1-13.
#' @keywords algebra
#' @example ./examples/ipca-examples.R
#' @export
ipca <- function (X,
                  ncomp = 2,
                  mode = "deflation",
                  fun = "logcosh",
                  scale = FALSE,
                  w.init = NULL,
                  max.iter = 200,
                  tol = 1e-04)
{
    #-- checking general input parameters --------------------------------------#
    #---------------------------------------------------------------------------#
    
    #-- X matrix
    if (is.data.frame(X))
        X = as.matrix(X)
    
    if (!is.matrix(X) || is.character(X))
        stop("'X' must be a numeric matrix.", call. = FALSE)
    
    if (any(apply(X, 1, is.infinite)))
        stop("infinite values in 'X'.", call. = FALSE)
    
    if (any(is.na(X)))
        stop("missing values in 'X'.", call. = FALSE)
    
    nc = ncol(X)
    nr = nrow(X)
    
    #-- put a names on the rows and columns of X --#
    X.names = colnames(X)
    if (is.null(X.names))
        X.names = paste("V", 1:nc, sep = "")
    
    ind.names = rownames(X)
    if (is.null(ind.names))
        ind.names = 1:nr
    
    #-- ncomp
    if (is.null(ncomp) ||
        !is.numeric(ncomp) || ncomp < 1 || !is.finite(ncomp))
        stop("invalid value for 'ncomp'.", call. = FALSE)
    
    ncomp = round(ncomp)
    
    if (ncomp > min(nc, nr))
        stop("use smaller 'ncomp'", call. = FALSE)
    
    #-- scale
    if (!is.logical(scale))
    {
        if (!is.numeric(scale) || (length(scale) != nc))
            stop(
                "'scale' should be either a logical value or a numeric vector of length equal to the number of columns of 'X'.",
                call. = FALSE
            )
    }
    
    X = scale(X, center = TRUE, scale = scale)
    sc = attr(X, "scaled:scale")
    
    if (any(sc == 0))
        stop("cannot rescale a constant/zero column to unit variance.",
             call. = FALSE)
    
    #-- mode
    choices = c("deflation", "parallel")
    mode = choices[pmatch(mode, choices)]
    
    if (is.na(mode))
        stop("'mode' should be one of 'deflation' or 'parallel'.", call. = FALSE)
    
    #-- fun
    choices = c("logcosh", "exp")
    fun = choices[pmatch(fun, choices)]
    
    if (is.na(fun))
        stop("'fun' should be one of 'logcosh' or 'exp'.", call. = FALSE)
    
    #-- w.init
    if (is.null(w.init))
    {
        w.init = matrix(1 / sqrt(ncomp), ncomp, ncomp)
    } else {
        if (!is.matrix(w.init) ||
            length(w.init) != (ncomp ^ 2) || !is.numeric(w.init))
            stop("'w.init' is not a numeric matrix or is the wrong size", call. = FALSE)
    }
    
    if (any(is.infinite(w.init)))
        stop("infinite values in 'w.init'.", call. = FALSE)
    
    if (sum(w.init == 0, na.rm = TRUE) == length(w.init))
        stop("'w.init' has to be a non-zero matrix", call. = FALSE)
    
    if (any(is.na(w.init)))
        stop("'w.init' has to be a numeric matrix matrix with non-NA values",
             call. = FALSE)
    
    
    #-- max.iter
    if (is.null(max.iter) ||
        !is.numeric(max.iter) || max.iter < 1 || !is.finite(max.iter))
        stop("invalid value for 'max.iter'.", call. = FALSE)
    
    max.iter = round(max.iter)
    
    #-- tol
    if (is.null(tol) ||
        !is.numeric(tol) || tol < 0 || !is.finite(tol))
        stop("invalid value for 'tol'.", call. = FALSE)
    
    
    #-- end checking --#
    #------------------#
    
    
    #-- ipca approach ----------------------------------------------------------#
    #---------------------------------------------------------------------------#
    
    V = svd(X)$v
    V = scale(V, center = TRUE, scale = TRUE)
    V = t(V)[1:ncomp, , drop = FALSE]
    
    if (mode == "deflation")
    {
        W = ica.def(V, ncomp = ncomp, tol = tol, fun = fun, alpha = 1,
                    max.iter = max.iter, verbose = FALSE, w.init = w.init)
    } else if (mode == "parallel") {
        W = ica.par(V, ncomp = ncomp, tol = tol, fun = fun, alpha = 1,
                    max.iter = max.iter, verbose = FALSE, w.init = w.init)
    }
    
    #-- independent loadings --#
    S = matrix(W %*% V, nrow = ncomp)
    
    #-- order independent loadings by kurtosis --#
    kurt = apply(S, 1, function(x) { n = length(x)
    x = x - mean(x)
    n * sum(x^4) / (sum(x^2)^2) - 3 } )
    ord = order(kurt, decreasing = TRUE)
    kurt = kurt[ord]
    S = S[ord, , drop = FALSE]
    norm = apply(S, 1, function(x) { crossprod(x) })
    S = t(sweep(S, 1, norm, "/"))
    
    #-- independent PCs / force orthonormality --#
    ipc = matrix(nrow = nr, ncol = ncomp)
    ipc[, 1] = X %*% S[, 1]
    ipc[, 1] = ipc[, 1] / sqrt(c(crossprod(ipc[, 1])))
    
    if (ncomp > 1)
    {
        for (h in 2:ncomp)
        {
            ipc[, h] = lsfit(y = X %*% S[, h], ipc[, 1:(h - 1)], intercept = FALSE)$res
            ipc[, h] = ipc[, h] / sqrt(c(crossprod(ipc[, h])))
        }
    }
    
    
    #-- output -----------------------------------------------------------------#
    #---------------------------------------------------------------------------#
    dimnames(S) = list(X.names, paste("IPC", 1:ncol(S), sep = ""))
    dimnames(ipc) = list(ind.names, paste("IPC", 1:ncol(ipc), sep = ""))
    
    cl = match.call()
    cl[[1]] = as.name('ipca')
    
    result = list(call = cl, X = X, ncomp = ncomp, x = ipc,
                  loadings=list(X=S),rotation=S,variates=list(X=ipc), kurtosis = kurt, unmixing = t(W),
                  mixing = t(t(W) %*% solve(W %*% t(W))),
                  names = list(var = X.names, sample = ind.names))
    
    class(result) = c("ipca", "pca")
    
    #calcul explained variance
    explX=explained_variance(X,result$variates$X,ncomp)
    result$prop_expl_var = list(X=explX)
    
    return(invisible(result))
}
mixOmicsTeam/mixOmics documentation built on Oct. 26, 2023, 6:48 a.m.