lfc direction of effect flipped for ashr, apeGLM, but accurate to biological data for non-shrunken
1
0
Entering edit mode
Ndimensional ▴ 20
@vlaufer-14169
Last seen 16 months ago
United States

Hello,

We have RNA-seq as well as functional data of various kinds.

For some of the most important genes in our study, we quantified protein and transcript levels using wet lab assays to confirm the size and direction of effect of the results.

Our original data were from microarray studies, these data showed massive downregulation of interferon responsive genes in the microarray data; subsequently, we have confirmed this in several different ways.

After running DESeq2 on the same samples, the results are completely congruent with prior studies. However, further processing the data with ashr or ApeGLM, as recommended by Michael Love elsewhere on this site and in published articles, the lfc direction flips.

this particular experiment has many more genes that are down than up. my understanding is that this is likely to affect results generated by Ashr, but not ApeGLM.

However, in either case, running lfcShrink produces values contrary to a great deal of accumulated experience.

This is addressed in the 2018 manuscript for apeGLM, but i cannot find suggestion therein why ApeGLM would behave in such a way in this case (as distinct from ashr, which does make sense granted the characteristics of the datasets). Any thoughts on this?

DESeq2 apeglm DESeq ApeGLM Ashr • 921 views
ADD COMMENT
0
Entering edit mode

You need to provide code, plots and some data examples for these genes to diagnose the issue. Has lfcShrink been run with an existing results object?

ADD REPLY
0
Entering edit mode
@mikelove
Last seen 16 hours ago
United States

the lfc direction flips

Do you mean the sign flips for all genes? I'm guessing this has to do with how the factors are coded.

Can you plot(ashr LFC, apeglm LFC)

ADD COMMENT
0
Entering edit mode

good guess. that was the first thing i checked.

i processed all 4 datasets using the same function, which selects appropriate columns, then renames them so they are standard across each dataset, then assigns them to the correct type (i.e., with as.factor or as.numeric), then assigns the same level as reference for all factors in each metadata table.

but, also, no, it isnt all genes. its some of them; others the sign stays the same between the initial DESeq2 results object and the apeGLM run.

VAL

ADD REPLY
0
Entering edit mode

For the ones which have changed sign, do they have a large LFC according to apeglm? If you use lfcThreshold=0.5 what is the svalue for these genes?

ADD REPLY

Login before adding your answer.

Traffic: 835 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6