Skip to contents

Wrapper function for parameters for the SCENIC binarisation

params_scenic_binarise.Rd

Each regulon gets its own threshold. A two-component Gaussian mixture is fitted and compared against a single Gaussian by BIC; if the mixture wins, the threshold is the kernel density minimum between the two component means, otherwise it falls back to mean + 2 * sd. This follows pySCENIC. AUCell fits six candidates and then lets the density trough override all of them whenever one exists, so the two land in much the same place. Turn bw_adjust up if shallow wobbles in the density are being picked up as troughs. AUCell effectively runs at 2.

Usage

params_scenic_binarise(bw_adjust = 1, n_grid = 512L, n_bins = 512L)

Arguments

bw_adjust

Numeric. Multiplier on the Silverman bandwidth of the kernel density estimate. Higher values smooth more. Defaults to 1.0.

n_grid

Integer. Number of points at which the density is evaluated between the two component means. Defaults to 512L.

n_bins

Integer. Number of histogram bins used to approximate the density. Defaults to 512L.

Value

A named list with the following elements:

  • bw_adjust - Numeric. Multiplier on the Silverman bandwidth of the kernel density estimate. Higher values smooth more. Defaults to 1.0.

  • n_grid - Integer. Number of points at which the density is evaluated between the two component means. Defaults to 512L.

  • n_bins - Integer. Number of histogram bins used to approximate the density. Defaults to 512L.

References

Aibar, et al., Nat Methods, 2017