summarizeFarmsExact2: Summarization Laplacian approach with exact computation

In cn.farms: cn.FARMS - factor analysis for copy number estimation

Description

This function implements an exact Laplace FARMS algorithm.

Usage

summarizeFarmsExact2(probes, mu = 1, weight = 0.5, weightSignal = 1,
  weightZ = 1, weightProbes = TRUE, cyc = c(10, 10), tol = 1e-05,
  weightType = "mean", centering = "median", rescale = FALSE,
  backscaleComputation = FALSE, maxIntensity = TRUE, refIdx, ...)

Arguments

probes	A matrix with numeric values.
mu	Hyperparameter value which allows to quantify different aspects of potential prior knowledge. Values near zero assumes that most positions do not contain a signal, and introduces a bias for loading matrix elements near zero. Default value is 0 and it's recommended not to change it.
weight	Hyperparameter value which determines the influence of the Gaussian prior of the loadings
weightSignal	Hyperparameter value on the signal.
weightZ	Hyperparameter value which determines how strong the Laplace prior of the factor should be at 0. Users should be aware, that a change of weightZ in comparison to the default parameter might also entail a need to change other parameters. Unexperienced users should not change weightZ.
weightProbes	Parameter (TRUE/FALSE), that determines, if the number of probes should additionally be considered in weight. If TRUE, weight will be modified.
cyc	Number of cycles. If the length is two, it is assumed, that a minimum and a maximum number of cycles is given. If the length is one, the value is interpreted as the exact number of cycles to be executed (minimum == maximum).
tol	States the termination tolerance if cyc[1]!=cyc[2]. Default is 0.00001.
weightType	Flag, that is used to summarize the probes of a sample.
centering	States how the data should be centered ("mean", "median"). Default is median.
rescale	Parameter (TRUE/FALSE), that determines, if moments in exact Laplace FARMS are rescaled in each iteration. Default is FALSE.
backscaleComputation	Parameter (TRUE/FALSE), that determines if the moments of hidden variables should be reestimated after rescaling the parameters.
maxIntensity	Parameter (TRUE/FALSE), that determines if the expectation value (=FALSE) or the maximum value (=TRUE) of p(z|x_i) should be used for an estimation of the hidden varaible.
refIdx	index or indices which are used for computation of the centering
...	Further parameters for expert users.

Value

A list including: the found parameters: lambda0, lambda1, Psi

the estimated factors: z (expectation), maxZ (maximum)

p: log-likelihood of the data given the found lambda0, lambda1, Psi (not the posterior likelihood that is optimized)

varzx: variances of the hidden variables given the data

KL: Kullback Leibler divergences between between posterior and prior distribution of the hidden variables

IC: Information Content considering the hidden variables and data

ICtransform: transformed Information Content

Case: Case for computation of a sample point (non-exception, special exception)

L1median: Median of the lambda vector components

intensity: back-computed summarized probeset values with mean correction

L_z: back-computed summarized probeset values without mean correction

rawCN: transformed values of L_z

SNR: some additional signal to noise ratio value

Author(s)

Andreas Mayr [e-mail] and Djork-Arne Clevert [e-mail] and Andreas Mitterecker [e-mail]

Examples

x <- matrix(rnorm(100, 11), 20, 5)
summarizeFarmsExact(x)

cn.farms documentation built on Nov. 8, 2020, 7:59 p.m.