NumericUtilities: Numeric Utilities
In seriph78/COTAN: COexpression Tables ANalysis

NumericUtilities

R Documentation

Numeric Utilities

Description

A set of function helper related to the statistical model underlying the COTAN package

Usage

funProbZero(dispersion, mu)

dispersionBisection(
  sumZeros,
  lambda,
  nu,
  threshold = 0.001,
  maxIterations = 100L
)

parallelDispersionBisection(
  genes,
  sumZeros,
  lambda,
  nu,
  threshold = 0.001,
  maxIterations = 100L
)

nuBisection(
  sumZeros,
  lambda,
  dispersion,
  initialGuess,
  threshold = 0.001,
  maxIterations = 100L
)

parallelNuBisection(
  cells,
  sumZeros,
  lambda,
  dispersion,
  initialGuess,
  threshold = 0.001,
  maxIterations = 100L
)

calcDist(data, method, diag = FALSE, upper = FALSE)

Arguments

`dispersion`	the estimated `dispersion` (a `n`-sized vector)
`mu`	the `lambda` times `nu` values (a `n \times m` matrix)
`sumZeros`	the number of genes not expressed in the relevant cell (a `m`-sized vector)
`lambda`	the estimated `lambda` (a `n`-sized vector)
`nu`	the estimated `nu` (a `m`-sized vector)
`threshold`	minimal solution precision
`maxIterations`	max number of iterations (avoids infinite loops)
`genes`	names of the relevant genes
`initialGuess`	the initial guess for `nu` (a `m`-sized vector)
`cells`	names of the relevant cells
`data`	a matrix or a data.frame of which we want to calculate the distance between columns
`method`	type of distance to use. Can be chosen among those supported by `parallelDist::parDist()`
`diag`	logical value indicating whether the diagonal of the distance matrix should be printed by print.dist.upper
`upper`	logical value indicating whether the upper triangle of the distance matrix should be printed by print.dist

Details

funProbZero is a private function that gives the probability that a sample gene's reads are zero, given the dispersion and mu parameters.

Using d for disp and \mu for mu, it returns: (1 + d \mu)^{-\frac{1}{d}} when d > 0 and \exp{((d - 1) \mu)} otherwise. The function is continuous in d = 0, increasing in d and decreasing in \mu. It returns 0 when d = -\infty or \mu = \infty. It returns 1 when \mu = 0.

dispersionBisection is a private function for the estimation of dispersion slot of a COTAN object via a bisection solver

The goal is to find a dispersion value that reduces to zero the difference between the number of estimated and counted zeros

parallelDispersionBisection is a private function invoked by estimateDispersionBisection() for the estimation of the dispersion slot of a COTAN object via a parallel bisection solver

The goal is to find a dispersion array that reduces to zero the difference between the number of estimated and counted zeros

nuBisection is a private function for the estimation of nu slot of a COTAN object via a bisection solver

The goal is to find a nu value that reduces to zero the difference between the number of estimated and counted zeros

parallelNuBisection is a private function invoked by estimateNuBisection() for the estimation of nu slot of a COTAN object via a parallel bisection solver

The goal is to find a nu array that reduces to zero the difference between the number of estimated and counted zeros

calcDist is a wrapper function that invokes parallelDist::parDist(): the main goal is to recover and finish the calculations via a fallback when there is a problem with the main algorithm