Can DESeq2 rlog normalized matrix reflects time course result directly?
1
0
Entering edit mode
pangtaihin • 0
@bd384ed5
Last seen 5 months ago
Hong Kong

Hello everyone :)

I have a three treatments (each treatment has 3 samples) gene expression matrix and my advisor wants me to do time course analysis with this data, while 3 treatments represents control, acute and chronic (0, 24h, 72h) respectively. I put the raw counts into R software ImpluseDE2 and get the time course result. Meanwhile, I also used DESeq2 to rlog normalize the same raw count, then directly compared the numerical size of the rlogged values of the mean of 3 samples each treatment. I got different result. In order to get time tendency result, is it OK to use DESeq2-rlogged raw counts to compare the numeric value directly? While the rlog values represent gene expression level.

However, I don't think it an easy job. If so, why so many developers develop R softwares specialized in time course analysis? ImpulseDE2, Mfuzz, MasigPro, EBSeq-HMM...

If rlog values cannot directly reflect the change tendency of gene expression level, why?

Thank you for your patience!

DESeq2 timecoursedata GeneExpression • 688 views
ADD COMMENT
1
Entering edit mode

If so, why so many developers develop R softwares specialized in time course analysis? ImpulseDE2, Mfuzz, MasigPro, EBSeq-HMM...

Briefly, because one usually assumes that time course experiments follow certain patterns. Like gradual increases, or sinusoidal rhythmicity. In contrast, a pattern like:

treatmentA: low-low-low-low-low-low
treatmentB: low-superHigh-low-middle-low-middle

...would probably be significant in something like a LRT test in DESeq2, but is no meaningful timecourse pattern. Specialized tools try to focus and identify meaningful patterns.

ADD REPLY
0
Entering edit mode

Thank you:)

ADD REPLY
1
Entering edit mode
@mikelove
Last seen 1 day ago
United States

If rlog values cannot directly reflect the change tendency of gene expression level, why?

In our workflow we recommend VST and rlog for visualization. For testing, you tend to have higher power working with raw counts and offsets. It matters less though when counts are high, and more when counts are close to zero.

ADD COMMENT
0
Entering edit mode

Thank you! But I apologize for cannot fully grasp your point. Do you mean that when the counts are high, rlogged matrix can represent the time course of genes?

ADD REPLY
1
Entering edit mode

High counts are generally less prone to noise compared to low counts. Bottom line is that you should stick to specialized testing frameworks such as DESeq2 (or edgeR, limma, or any specialized time-course framework) rather than trying to stitch testing together yourself.

ADD REPLY
0
Entering edit mode

Thank you very much ATpoint, but DESeq2, edgeR, limma are not specialized time-course framework, I think we might seek for another framework.

ADD REPLY
0
Entering edit mode

You can try other frameworks, but as I said, three timepoints are hardly a timecourse. Some clever combinations of filtering your DEGs for patterns that make sense and/or choosing patterns on a clustered heatmap might do just as well.

ADD REPLY
0
Entering edit mode

Yes, ATpoint. I not only used specialized time course frameworks but also used DESeq2 rlog heatmap to identify the time course clusters. But they have different resluts. How should I put my results into perspective? How to verify and which one should I use?

Thank you again for your generous help, ATpoint:)

ADD REPLY
1
Entering edit mode

For hands-on guidance consult a local collaborator. This is outside of what can be done here with a few sentences.

ADD REPLY

Login before adding your answer.

Traffic: 416 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