Choose an Optimal Number of Signatures — ChooseNumberSignatures • utanos

Uses the NMF package to estimate the rank of the matrix provided. In the context of signature finding, this is equivalent to determining the optimal number of signatures.

Usage

ChooseNumberSignatures(
  sample_by_component,
  min_sig = 3,
  max_sig = 12,
  iter = 100,
  cores = 4,
  nmfalg = "brunet",
  seed = 77777
)

Arguments

sample_by_component: Matrix. Expects a sum-of-posteriors probability matrix. Likely the output from GenerateSampleByComponentMatrix().
min_sig: Integer. Minimum number of signatures to consider.
max_sig: Integer. Maximum number of signatures to consider.
iter: Integer. The number of runs of NMF to perform at each rank with a random seed.
cores: Integer. (NMF param) The number of cores to use when running NMF.

Value

A figure. A figure plotting the cophenetic distance, the dispersion, sparseness and silhouette score.

Details

The NMF Package (Brunet algorithm specification) is used to deconvolute the sample-by-component sum-of-posteriors matrix into a sample-by-signature matrix and a signature-by-component matrix. By default a search interval of 3-12 is used for the rank. To compare/for context the input matrix is also shuffled and the same procedure performed. The results are co-plotted to give a baseline.