I am trying to compare metagenomics datasets without biological replicates. (Not ideal, i know. But biological replication is not possible for the communities we want to learn about.) Right now I'm working with normalized count data, but I've got raw reads and am happy to recompute anything.
I would like to rank features by how much their abundance decreases or increases between samples. Basically i want log2 fold changes. But those get too unreliable when read counts for a feature are low/zero. GFOLD looks like what i need, but the tool is not available anymore.
Do you have any recommendations regarding either a specific tool or relevant texts? I'm not set on using fold changes, open to basically anything that helps with comparing the datasets.
*edit: Thanks to the corresponding authors i found GFOLD at a new URL.
This seems like a typical MA plot scenario where you would plot
M = log2(a/b)on the y-axis andA = log2(sqrt(a*b))on the x-axis, and you might see something resembling a sideways Christmas Tree. You would be examining ratio as a function of general abundance, and it allows you to see what you're talking about - easy to get ratios from differences in small numbers and that trend generally decreases as the numbers get larger. Without stats, one thing you cold do would be to favor ratios that take more reads into account, over those that take fewer reads. Thus you could rank by some variant of abs(M) * A. You'd be drilling into the data from a diagonal based on the M and A values. Of course, this approach is a heuristic based on the shape of the data, and having not seen your data it may not be valid, but it is a way to rank taking ratio and abundance into account.Thank you, i like that idea and will try how it looks in practice!