MIDASim (version 2.0)

MIDASim.modify: Modifying MIDASim model

Description

MIDASim.modify() modifies the fitted MIDASim.setup model according to user specification that one or multiple of the following characteristics, such as the library sizes, taxa relative abundances, location parameters of the parametric model can be changed. This is useful if the users wants to introduce an 'effect' in simulation studies.

Usage

MIDASim.modify(
  fitted,
  lib.size = NULL,
  mean.rel.abund = NULL,
  gengamma.mu = NULL,
  sample.1.prop = NULL,
  taxa.1.prop = NULL,
  individual.rel.abund = NULL,
  ...
)

Value

Returns an updated list with different elements depending on the value of fitted$mode:

n.sample

Target sample size in the simulation.

lib.size

Target library sizes in the simulation.

taxa.1.prop

Updated proportions of non-zero values for each taxon.

sample.1.prop

Updated proportion of non-zero cells for each subject.

theta

Mean values of the multivariate normal distribution in generating presence-absence data.

eta

Adjustment to be applied to samples in generating presence- absence data.

Arguments

fitted

Output from MIDASim.setup.

lib.size

Numeric vector of pre-specified library sizes (length should be equal to n.sample if specified). In nonparametric mode, if lib.size is specified, both taxa.1.prop and sample.1.prop should be specified.

mean.rel.abund

Numeric vector of specified mean relative abundances for taxa. Length should be equal to n.taxa in fitted.

gengamma.mu

Numeric vector of specified location parameters for the parametric model (generalized gamma model). Specify either mean.rel.abund or gengamma.mu, not both. Length should be equal to n.taxa in fitted. See Details. This argument is only applicable in parametric mode.

sample.1.prop

Numeric vector of specified proportion of non-zeros for subjects (the length should be equal to n.sample in fitted). This argument is only applicable in nonparametric mode.

taxa.1.prop

Numeric vector of specified proportion of non-zeros for taxa (the length should be equal to n.taxa in fitted). This argument is only applicable in nonparametric mode.

individual.rel.abund

Numeric matrix of expected relative abundances with n.sample rows and n.taxa columns (rows should sum to 1). Provides subject‑specific mean compositions and therefore overrides mean.rel.abund and gengamma.mu. Only applicable in parametric mode.

...

Additional arguments. If SCAM model is chosen for parameter changes under the non-parametric mode, specify SCAM = T.

Author

Mengyu He

Details

The parametric model in MIDASim is a location-scale model, specifically, a generalized gamma model for relative abundances \(\pi\) of a taxon. Denote \(t = 1/\pi\). The generalized gamma distribution for \(t\) is chosen so that

$$ln(t)\ =\ \mu\ +\ \sigma \cdot w$$

where \(w\) follows a log gamma distribution with a shape parameter \(1/Q\). MIDASim fits the model to the template data and estimates parameters \(\mu\), \(\sigma\) and \(Q\) by matching the first two moments of \(\pi\) and maximizing the likelihood.

Examples

# \donttest{
  data("throat.otu.tab")
  otu.tab = throat.otu.tab[,colSums(throat.otu.tab>0)>1]

  fitted = MIDASim.setup(otu.tab, mode = 'parametric')

  # modify library sizes
  fitted.modified <- MIDASim.modify(fitted,
                                    lib.size = sample(fitted$lib.size, 2*nrow(otu.tab),
                                                    replace = TRUE) )

  # modify mean relative abundances
  fitted.modified <- MIDASim.modify(fitted,
                                    mean.rel.abund = fitted$mean.rel.abund * runif(fitted$n.taxa))

# }