NBumi_HVG: Variance-based Feature Selection Accounting for Library Size...

NBumiHVGR Documentation

Variance-based Feature Selection Accounting for Library Size and Sample Variance

Description

Tests for significantly high variability in droplet-based datasets.

Usage

	NBumiHVG(counts, fit, fdr_thresh=0.05, suppress.plot=FALSE, method=c("DANB", "basic"))

Arguments

counts

raw count matrix (e.g. from NBumiConvertData).

fit

output from NBumiFitModel or NBumiFitBasicModel.

fdr_thresh

multiple testing correction threshold to apply to filter output.

suppress.plot

whether to plot mean vs variance & selected features.

method

whether to use DANB dispersions or raw sample variances.

Details

Assumes a constant dispersion parameter due to technical noise (see: [1]), which is estimated using linear regression. Gene-specific observed sample variances are compared to their respective expected variances. Significance is evaluated using a Z-test with the expected variance of the sample variance for a Negative Binomial (see: [2]).

The method argument controls whether the expected and observed variances are adjusted to account for uneven library sizes between cells. The default "DANB" option does the adjustment.

Value

a data.frame with columns: Gene, effect.size, p.value, q.value

References

[1] Svensson, V. (2019) Droplet scRNA-seq is not zero-inflated. bioRxiv. doi: https://doi.org/10.1101/582064 [2] Rose, C. and Smith, M. D. Mathematical Statistics with Mathematica. New York: Springer-Verlag, 2002. p264

Examples

	library(M3DExampleData)
	counts <- NBumiConvertData(Mmus_example_list$data)
	fit <- NBumiFitModel(counts);
	HVGs <- NBumiHVG(counts, fit, suppress.plot=TRUE);
	HVGs_uncorrected <- NBumiHVG(counts, fit, suppress.plot=TRUE, method="basic");

tallulandrews/M3Drop documentation built on March 6, 2024, 1:49 a.m.