estNPresp: Estimate the taxon-wise response functions non-parametrically

In RCM: Fit row-column association models with the negative binomial distribution for the microbiome

Description

Estimate the taxon-wise response functions non-parametrically

Usage

estNPresp(
  sampleScore,
  muMarg,
  X,
  ncols,
  thetas,
  n,
  coefInit,
  coefInitOverall,
  dfSpline,
  vgamMaxit,
  degree,
  verbose,
  allowMissingness,
  naId,
  ...
)

Arguments

sampleScore	a vector of length n with environmental scores
muMarg	the offset matrix
X	the n-by-p data matrix
ncols	an integer, the number of columns of X
thetas	a vector of length p with dispersion parameters
n	an integer, the number of samples
coefInit	a 2-by-p matrix with current taxon-wise parameter estimates
coefInitOverall	a vector of length 2 with current overall parameters
dfSpline	a scalar, the degrees of freedom for the smoothing spline.
vgamMaxit	Maximal number of iterations in the fitting of the GAM model
degree	The degree if the parametric fit if the VGAM fit fails
verbose	a boolean, should number of failed fits be reported
allowMissingness	A boolean, are missing values present
naId	The numeric index of the missing values in X
...	further arguments, passed on to the VGAM:::vgam() function The negative binomial likelihood is still maximized, but now the response function is a non-parametric one. To avoid a perfect fit and overly flexible functions, we enforce smoothness restrictions. In practice we use a generalized additive model (GAM), i.e. with splines. The same fitting procedure is carried out ignoring species labels. We do not normalize the parameters related to the splines: the psis can be calculated afterwards.

Value

A list with components

taxonCoef	The fitted coefficients of the sample-wise response curves
splinesList	A list of all the B-spline objects
rowMar	The row matrix
overall	The overall fit ignoring taxon labels, as a list of coefficients and a spline
rowVecOverall	The overall row vector, ignoring taxon labels

RCM documentation built on Nov. 8, 2020, 5:22 p.m.