Skip to contents

The closed-form alternative to scTransform: one shared dispersion instead of a fitted model per gene. Much cheaper, and on most data sets it ranks genes about as well.

Usage

params_sc_apr(theta = 100, min_cells = 5L, clip_min = NULL, clip_max = NULL)

Arguments

theta

Numeric. The shared negative binomial dispersion. Inf gives the Poisson limit. Defaults to 100.0.

min_cells

Integer. Minimum number of cells a gene must be detected in to be retained. 0L keeps everything. Defaults to 5L.

clip_min

Numeric or NULL. Lower residual clipping bound. NULL uses -sqrt(n_cells). Must be given together with clip_max. Defaults to NULL.

clip_max

Numeric or NULL. Upper residual clipping bound. NULL uses sqrt(n_cells). Must be given together with clip_min. Defaults to NULL.

Value

A named list with the following elements:

  • theta - Numeric. The shared negative binomial dispersion. Inf gives the Poisson limit. Defaults to 100.0.

  • min_cells - Integer. Minimum number of cells a gene must be detected in to be retained. 0L keeps everything. Defaults to 5L.

  • clip_min - Numeric or NULL. Lower residual clipping bound. NULL uses -sqrt(n_cells). Must be given together with clip_max. Defaults to NULL.

  • clip_max - Numeric or NULL. Upper residual clipping bound. NULL uses sqrt(n_cells). Must be given together with clip_min. Defaults to NULL.

References

Lause, Berens and Kobak, Genome Biology, 2021.