meanLogX: Computes the mean of log-scaled values

In BaderLab/scClustViz: Differential Expression-based scRNAseq Cluster Assessment and Viewing

Computes the mean of log-scaled values

Description

Computes the arithmetic mean in linear space of log-scaled values, and returns the result in the same log scale.

Usage

meanLogX(data, ncell, ex = 2, pc = 1)

Arguments

data	A numeric vector in log scale for which the arithmetic mean in linear scale is to be calculated and returned in the same log scale.
ncell	The number of samples (cells assuming an scRNAseq experiment) in the data. This is not necessarily the same as the length of data. Since its inverse will be used as the pseudocount value for all calculations, it should be consistent for all calls of mean.logX.
ex	The log base of the data. Traditionally gene expression data is represented in base 2, although some methods (ie. Seurat's normalization scheme) use the natural log. (default=2)
pc	The pseudocount used when converting the data to log-scale initially. (default=1)

Details

Generally a pseudocount of 1 is added to log-scaled values to prevent +/-Inf results. However, adding a pseudocount of 1 to the log-scaled mean prior to gene expression ratio calculations in log space skews the result quite dramatically, so instead we add a small pseudocount to avoid +/- inf results when means are zero, without the same skewing. Adding a very small (ie 1e-99) number means that means of zero get set to a large negative log-mean, when it might be more appropriate to have those values fall closer to the smallest non-zero log-mean. By using a pseudocount of 1 / number of samples in the experiment, we ensure that log(zero) is smaller than any non-zero log-mean, while still being in the same ballpark.

BaderLab/scClustViz documentation built on Sept. 10, 2023, 11:51 p.m.