Question

metagenomeSeq 1.16.0: unique features give NA's for logFC / p-val?

0

Entering edit mode

handibles ▴ 10

@handibles-12652

Last seen 6.8 years ago

Hey Devs,

Thanks for the package and it's upkeep. I've hit a snag comparing two conditions (A=6 samples, B=9 samples), wherein feature counts in A = 0 or 1, counts in B =100 - 15,000, but the logFC and p-values are NA/NaN's. +samples typically are all of one group, 1-0 of the other.

I would expect 1:15000 to be a significant signal, and can only think of one thing, the paragraph in the metagenome.pdf, which states

_"Warning: The user should restrict significant features to those with a minimum number of positive samples. What this means is that one should not claim features are significant unless the effective number of samples is above a particular percentage. For example, fold-change estimates might be unreliable if an entire group does not have a positive count for the feature in question."_

is this what I'm seeing with this data? Can produce MRE at a later point if necessary.

Thanks for the assist

example:

`NA`	`+samples in group 0`	`+samples in group 1`	`counts in group 0`	`counts in group 1`	`logFC`	`se`	`pvalues`	`adjPvalues`
`AB669249`	`1`	`7`	`1`	`15584`	`NA`	`NA`	`NaN`	`NaN`
`JN052751.`	`0`	`9`	`0`	`4443`	`NA`	`NA`	`NaN`	`NaN`
`AB2745058`	`1`	`8`	`1`	`4311`	`NA`	`NA`	`NaN`	`NaN`
`AB274517`	`0`	`9`	`0`	`2881`	`NA`	`NA`	`NaN`	`NaN`
`CU918909`	`0`	`6`	`0`	`2643`	`NA`	`NA`	`NaN`	`NaN`
`FN56316`	`0`	`9`	`0`	`2240`	`NA`	`NA`	`NaN`	`NaN`
`CU92752`	`0`	`8`	`0`	`2023`	`NA`	`NA`	`NaN`	`NaN`
`KM67594`	`0`	`9`	`0`	`1645`	`NA`	`NA`	`NaN`	`NaN`
`AJ50619`	`1`	`9`	`1`	`1262`	`NA`	`NA`	`NaN`	`NaN`
`CU9189`	`0`	`6`	`0`	`1236`	`NA`	`NA`	`NaN`	`NaN`

metagenomeseq • 1.1k views

ADD COMMENT • link updated 4 months ago by liqiong464 • 0 • written 7.1 years ago by handibles ▴ 10

score 0 · Answer 1 · 2017-03-27

0

Entering edit mode

Joseph Nathaniel Paulson ▴ 280

@joseph-nathaniel-paulson-6442

Last seen 7.1 years ago

United States

The fitFeatureModel will return NA p-values for any features where there are less than 3 positive samples in a group. The full model is not estimable otherwise.

Will modify this later once I have a bit more time.

ADD COMMENT • link 7.1 years ago Joseph Nathaniel Paulson ▴ 280

0

Entering edit mode

Hey,

I encountered similiar problems now as handibles metentioned. But it seems a bit different somehow.
I have 51 samples divided into 3 groups, thus 17 samples per group. I filtered features present <0.5 samples per group. Then used the The fitFeatureModel to do comparisons between 2 groups. But I got many NAs in columns se, pvalue and adjPvalues.

Below is the code: sIE_Stable_mod <- model.matrix(~1+ Group, data = pData(sIEStable_meta_filt_css)) sIE_Stable_res1 <- fitFeatureModel(sIEStable_meta_filt_css, sIE_Stable_mod) sIE_Stable_res1_coef<-MRcoefs(sIE_Stable_res1,number=nrow(MRcounts(sIEStable_meta_filt_css))) enter image description here
Can i use the results directly or something goes wrong?

Many thanks

ADD REPLY • link 4 months ago liqiong464 • 0