voila tsv output error

26 views
Skip to first unread message

Sarah Djeddi

unread,
Feb 22, 2021, 9:47:14 AMFeb 22
to majiq_voila
Hello all,
I am trying to parse results from deltapsi output and to obtain same results as in voila tsv or voila view, but I have different results.
My question:
How can I parse deltapsi output? Should I consider E(dPSI) and P dPSI ? Or only one the two values? I don't understand how is it done. 

Thank you,

Sarah

Matthew Gazzara

unread,
Feb 24, 2021, 1:44:51 PMFeb 24
to majiq_voila
Hi Sarah,

Thank you for your interest. To get a bit more information in order to help you out, can you share the command you used to generate the voila tsv? 

As for your question about parsing / filtering your deltapsi (dPSI) output. Roughly E(dPSI) will tell you the expected magnitude of the change between your two groups while P(|dPSI| > threshold) will tell you the probability that a junction in an LSV changed greater than that threshold.

These two are related, but P(|dPSI|>X) > 0.95 would be what we consider a "high-confidence" change (we typically use X=10% or X=20%). This means that 95% of the posterior distribution of dPSI between your compared groups sits beyond your threshold (X) value for dPSI. 

LSVs with higher E(dPSI) values will tend to have a higher probability of change, but high read coverage can also drive up this probability higher (we are more sure the dPSI between the groups is "real" (larger than your threshold)). However, a higher probability here does not necessarily mean a larger E(dPSI). With many reads we can be very confident there is a change of at least 20% (P(|dPSI|>20%) ~ 1), without having the highest expected PSI for that LSV (E(dPSI) can be ~20%) out of all LSVs analyzed. 

Hopefully this gives you a bit more intuition on how to use these and don't hesitate to ask more questions.  

Thanks!
Matt 
Reply all
Reply to author
Forward
0 new messages