Skip to contents

corncob analysis

Source: R/corncob.R
step_corncob.Rd

Corncob an individual taxon regression model that uses abundance tables and sample data. corncob is able to model differential abundance and differential variability, and addresses each of the challenges presented below:

Usage

step_corncob(
  rec,
  phi.formula = stats::formula(~1),
  formula_null = NULL,
  phi.formula_null = NULL,
  link = "logit",
  phi.link = "logit",
  test = "Wald",
  boot = FALSE,
  B = 1000,
  filter_discriminant = TRUE,
  fdr_cutoff = 0.05,
  fdr = "fdr",
  log2FC = 0,
  rarefy = FALSE,
  id = rand_id("corncob"),
  engine_args = list()
)

Arguments

rec

A Recipe object. The step will be added to the sequence of operations for this Recipe. Execution requires add_model() before prep(); the step can be configured before adding the model.

phi.formula

An object of class formula without the response: a symbolic description of the model to be fitted to the dispersion.

formula_null

Retained design argument, default NULL. Explicit values are overridden with a warning by the add_model() contrasts.

phi.formula_null

Formula for overdispersion under null, without response. Default NULL. Explicit values are overridden with a warning by the add_model() contrasts.

Link function for abundance covariates, defaults to "logit".

Link function for dispersion covariates, defaults to "logit".

test

Hypothesis test, default "Wald". Only Wald contrasts are supported; other tests make the step incompatible with the shared model.

boot

Whether to request bootstrap inference, default FALSE. Bootstrap is unsupported and makes the step incompatible.

B

Retained bootstrap iteration setting. Not used for Wald contrasts.

filter_discriminant

Boolean. Defaults to TRUE. If FALSE, discriminant taxa will not be filtered out.

fdr_cutoff

Integer. Defaults to 0.05. Desired type 1 error rate.

fdr

Character. Defaults to "fdr". False discovery rate control method, see p.adjust for more options.

log2FC

Minimum absolute effect on the reported coefficient scale. Despite the argument name, a logit-link coefficient is not a log2 fold change. Default zero.

rarefy

Boolean indicating if OTU counts must be rarefyed. This rarefaction uses the standard R sample function to resample from the abundance values in the otu_table component of the first argument, physeq. Often one of the major goals of this procedure is to achieve parity in total number of counts between samples, as an alternative to other formal normalization procedures, which is why a single value for the sample.size is expected. If 'no_seed', rarefaction is performed without a set seed.

id

Unique, non-empty identifier for this step. DA step IDs must not use the reserved internal names taxa_id, contrast_id, comparison, contrast_type, var, effect, sum, .tax_order, .contrast_order or .effect_order. Other identifiers, including names such as .custom, are retained unchanged.

engine_args

Named lists of advanced arguments for the native fit stage. Arguments managed by dar or exposed above cannot be overridden.

Value

An object of class Recipe

Details

  • different sequencing depth

  • excessive zeros from unobserved taxa

  • high variability of empirical relative abundances (overdispersion)

  • within-taxon correlation

  • hypothesis testing with categorical and continuous covariates

Centralized models use non-bootstrap Wald contrasts. With engine_args = list(fit = list(robust = TRUE)), inference uses corncob's native sandwich covariance rather than the inverse Hessian. This does not add support for random effects or repeated-measures dependence.

See also

Other Diff taxa steps: step_aldex2(), step_aldex3(), step_ancom(), step_deseq(), step_lefse(), step_linda(), step_maaslin(), step_rademu(), step_wilcox()

Examples

data(metaHIV_phy)

rec <- recipe(metaHIV_phy) |>
  add_model(~ RiskGroup2, targets = "RiskGroup2", tax_level = "Phylum")

rec_default <- step_corncob(rec)

# Use a stricter significance threshold.
rec_strict <- step_corncob(rec, fdr_cutoff = 0.01)