appear to be constant with groups

[celinechen1885]

Hi!

I'm working with a shotgun metagenomics data, and we aim to analyze the functional profile.
we have 224 samples in total, with 356 different pathways.
below are the codes that I used
metagenomic_lefse <- read.csv
se <- SummarizedExperiment(assays = list(counts = as.matrix(metagenomic_lefse)))
colData(se) <- DataFrame(metadata)
se$group <- as.factor(se$group)
se$DI <- as.factor(se$DI)
se_relab <- relativeAb(se)

res <- lefser(se_relab, groupCol = "group")

I got an error message:
Error in lda.default(x, grouping, ...) :
variables 85 145 appear to be constant within groups

Could anyone help me with it?
Thanks!

[lwaldron]

This seems to happen with normalization to column sums adding to 1 instead of 1 million as in the original implementation. The following code after normalization accomplishes this, and we will make this the default of the relativeAb function.

assay(rse) = assay(rse)*1e6 # will become the default of relativeAb

[lwaldron]

@Peacesandy append * 1e6 to the line

lefser/R/utils.R

Line 28 in db8f36a

res <- assay_data / div

.

The following code currently reproduces the problem, so use it to create a testthat unit test of the fix:

library(lefser)
data(zeller14)
zeller14sub <- zeller14[, zeller14$study_condition != "adenoma"]
zeller14ra <- relativeAb(zeller14sub)
lefser(zeller14ra, groupCol = "study_condition")

[Peacesandy]

Okay, I'll make a pull request as soon as I'm done

[Peacesandy]

I just made a pull request
@lwaldron

#34

[lwaldron]

Fixed by 8864568 (v1.13.2) and this error no longer appears. Thank you for the report, @celinechen1885 . @asyakhl will you push to bioc-devel ASAP? @celinechen1885, the easiest way to access this fix for now is to install from GitHub (BiocManager::install("waldronlab/lefser")), or just multiply your relative abundance values by 1e6.

[lwaldron]

@asyakhl please push to bioc-devel and cherry-pick to bioc-release. This is an important bugfix.

[LiNk-NY]

@lwaldron I've updated both. Thanks!

[lwaldron]

Thanks Marcel!