Entering edit mode
Angel Liang
•
0
@angel-liang-23540
Last seen 3.8 years ago
Hi,
I have results from using the DEXSeqDataSetFromHTSeq() function in DEXSeq. I want to calculate exon PSI (given each transcript has its own groupID and each exon has its own featureID), can I just convert the exon usage coefficients to counts by taking 2^x then do (ExonCount/SumOfAllExonCounts)?
I am aware that most approaches just use the raw counts, but I am avoiding it because that doesn't seem to take advantage of DEXSeq's statistical model, I think?
Also, would it be more accurate to calculate a delta PSI value using the log2fold value? Is there any easy/quick way I can achieve this?
thanks
Hi Angel,
Thanks for your message. I have not done or benchmark the approaches to calculate PSIs that you are suggesting so I can't unfortunately comment on accuracy, etc. However, I'm not sure using DEXSeq's beta estimates or log fold changes is the best way to do that. If I were to calculate PSIs, I'd rather use raw counts (counts in exon/tot counts in gene) and the resulting values should somewhat reflect changes detected by DEXSeq. Another point to consider is that PSI's are commonly used to quantify local splicing events (e.g. exclusion vs inclusion reads of a cassette exon) rather than using counts per exon.
May I ask what is the goal of your analysis?
Alejandro
Hi Alejandro,
I am attempting to quantify differential alternative splicing events by feeding junction-spanning counts into DEXSeq. I have each featureID as an exon, each containing 2 groupIDs: exclusion reads and inclusion reads for the exon. Each exon can be assumed to originate from exactly one Local Isoform Variant each.
I am not sure if you are familiar with a tool called "JUM" (Wang, Q. and Rio, D.C., 2018. JUM is a computational method for comprehensive annotation-free analysis of alternative pre-mRNA splicing patterns. Proceedings of the National Academy of Sciences, 115(35), pp.E8181-E8190.)
It uses DEXSeq to detect differential junction usage and I am trying to do something similar here, except that they used the raw counts to calculate PSI hence deltaPSI.
I was thinking that using the ExonCount values from DEXSeq would be more accurate, since they are the fitted values? Please correct me if I am wrong - I am not entirely sure about this, which is why I was hoping you could address this issue
Regards
Thanks for the clear explanation, now I understand better your analysis. For the PSI calculations, I would recommend to use raw counts. I don't see how you can use the beta coefficients from DEXSeq to calculate PSIs.
In your analysis, which uses inclusion vs exclusion reads, the delta PSIs and DEXSeq's log2 fold changes are measuring the same thing, so these metrics should be correlated.
thanks alejandro, that is very helpful :)