Question: Differential transcript expression (DTE) analysis
0
gravatar for rbenel
9 months ago by
rbenel0
rbenel0 wrote:

I was going through the workflow "Swimming downstream: statistical analysis of differential transcript usage following Salmon quantification" which can be found here, https://f1000research.com/articles/7-952/v3#f4 and I wanted to clarify something.

Under the section DTE evaluation there is a sentence that says "For this assessment, all of the simulated non-null transcripts count as true positives of differential transcript expression, whether they originate from the simulated “DGE”, “DTE”, or “DTU” genes. For most of the methods, we simply provided the transcript-level data to the same functions as for the DGE analysis."

As mentioned there are investigators that prefer to assess DE on a per-transcript basis, does this mean one could use DGE methods at the transcript level?Is DTE in essence capturing any transcript that is DE, regardless of that status of that transcript (i.e. how many transcripts are expressed, and if the total gene-level does not change)?

Thanks!

dtu rnaseqdtu • 332 views
ADD COMMENTlink modified 9 months ago by Michael Love25k • written 9 months ago by rbenel0

As a follow up, here is an example of a gene that is DGE following DESeq2, and also has two transcripts that are DTU following DEXseq and stageR analysis with a significant adjusted p-value at the level of the gene and transcript.

bargraph: https://www.dropbox.com/s/z95ri9keyt48bcw/example_bioconducter.png?dl=0

ADD REPLYlink modified 9 months ago • written 9 months ago by rbenel0
Answer: Differential transcript expression (DTE) analysis
1
gravatar for Michael Love
9 months ago by
Michael Love25k
United States
Michael Love25k wrote:

hi,

We showed in that paper that the DGE methods can in practice be applied to transcript level quantification, and that some of the DGE methods perform reasonable well at transcript level on that simulated dataset which had various types of DTE spiked in, and realistic technical biases on the coverage. You can read off the performance at various sample sizes. As we say, I find that looking at DGE and DTU to be more interpretable about the changes in transcription (e.g. the DTU vs. DGE scatterplots we have in the workflow and paper), but I think it depends on the experiment and the biological question of interest which approach you might take. For the DTE analysis you are getting a p-value for DE of each transcript.

ADD COMMENTlink written 9 months ago by Michael Love25k

Just to clarify, "For the DTE analysis you are getting a p-value for DE of each transcript" means if you use a DGE method on transcript level data, you should be getting a p-value for every transcript that is DE?

In addition, I have a question regarding the link I posted with an example, it seems clear from the boxplot that the gene is DE, it also looks like the individual transcripts (for the sure the higher expressed one) is DE, but it seems to me the total expression of the transcripts is changing, so I am wondering why it is considered DTU. From what I understand, DTU should be capturing only transcripts whose total gene level does not change, but the individual transcripts are DE across a condition, time etc...

Thank You!

ADD REPLYlink modified 9 months ago • written 9 months ago by rbenel0

With DTE, you get a pvalue for every transcript. Small pvalues suggest rejection of the null.

DTU is when the isoform proportions change. It is possible to have this happen in absence of DGE but also possible to have both at the same time. Again, see the DTU vs DGE scatterplot I posted above. The points in the middle have evidence of both DTU and DGE because the isoform proportions within change as well as the total gene output.

ADD REPLYlink modified 9 months ago • written 9 months ago by Michael Love25k

OK.

I understand that the middle area can be both DGE and DTU. In the specific example, I was concerned about the evidence for DTU, it doesn't seem to me that there is a large change in isoform proportions, that's all.

ADD REPLYlink written 9 months ago by rbenel0

But there is a clear change in proportion (if I’m reading the plot correctly) from day 0 to day 1. The lower expressed isoform increases while the major isoform decreases.

ADD REPLYlink written 9 months ago by Michael Love25k

You are reading it correctly, and yes there is a change in the proportion, but the global picture suggests that the major isoform is decreasing with time (regardless of its relationship to the minor isoform). Therefore, I am not sure that the increase in the minor isoform between day 0 and day 1 is in order to "compensate" for the decrease in the major isoform....

Perhaps I am mistakenly trying to interpret both biological significance as well as statistical significance...

ADD REPLYlink modified 9 months ago • written 9 months ago by rbenel0

All I can say is that this data suggests a change in total expression and clear changes in proportion as well. If you want, eg a total switch in the major isoform you could do some posthoc filtering (I don’t have code for this).

ADD REPLYlink written 9 months ago by Michael Love25k

Yes, a total switch would be the most biologically interesting in this case... If something comes of the analysis I can post the code here in case it interests anyone else.

Thanks Michael for all of the help :)

ADD REPLYlink written 9 months ago by rbenel0
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 16.09
Traffic: 124 users visited in the last hour