RMT: Performs the RMT analysis on a gene similarity matrix.

In SystemsGenetics/KINC.R: Provides additional functions for network files created using KINC.

Description

During Random Matrix Theory (RMT) analysis, a subset of the similarity matrix is tested to see how closely the nearest neighbor spacing distribution (NNSD) of its eigenvalues. A NNSD that appears Guassian is indicative of a random matrix. Naturally occuring networks do not exhibit random behavior and their NNSD appears Poisson. This function iterates through a successively decreasing similarity score values, creates a subset of the similarity matrix containing only genes with similarity values at or above the test score and performs a Chi-square test on its NNSD. Prior to testing, eigenvalues are spectrally unfolded and then fit with a spline. The NNSD is essentially a historgram with 60 bins. Therefore, the Chi-square test has 60 degrees of freedom and the test statistic, 99.607, corresponding to a p-value of 0.001. Thus A Chi-square test less than or equal to 99.607 indicates that the similarity submatrix appears Poisson.

Usage

RMT(
  net,
  start = 0.99,
  end = 0.75,
  step = -0.001,
  min.eigens = 100,
  save.plot = FALSE
)

Arguments

net	A network dataframe containing the KINC-produced network. The loadNetwork function imports a dataframe in the correct format for this function.
start	The similarity score at which to start RMT analysis. The RMT will iterate through successively lower similarity scores. By default it performs the ChiSquare test of a reduced matrix that only contains genes with correlation at or above this start value. Default is 0.99.
end	The similarity score at which the RMT analysis will not proceed further. Default is 0.75
step	The size that the similarity score is reduced at each iteration tested. Default is -0.001.
min.eigens	To perform a Chi-squared test of the NNSD, the eigenvalues of the reduces similarity matrix must first be calculated. This argument sets the limit for the number of eigenvalues that must be present in order to perform the test. This roughly corresponds to the number of genes in the reduced similarity matrix. The actual number of eigenvalues may be smaller than the number of genes if duplicates are present. Duplicates are removed. Default is 100.
save.plot	Set to TRUE to save a plot at each iteration.

SystemsGenetics/KINC.R documentation built on Nov. 10, 2021, 9:22 p.m.