Some methods to filter the results

[MicSonn]

Dear IntaRNA developers,
I'm examining a set of miRNAs and their interactions between 5'UTRs, 3'UTRs, and CDSs of human genes.

For computational reasons, I already applied filters when launching intaRNA. Examples: --seedMaxE=-10, --outMaxE=-26, --seedMinPu=0.001 --outMinPu=0.001 --outDeltaE=100, and helixMinPu=0.7.

Do you have any suggestions on how to filter the results based on the output parameters?
--outCsvCols '*' --mode=M --model=P --acc=C --outMode=C

A good filtering/sorting approach might be to sort increasing E (overall interaction energy) AND decreasing P_E (probability of an interaction (site) within the considered ensemble).

Which P_E values ​​should I discard?

Thank you.

Best,
MicSonn!

[martin-raden]

Hi MicSonn,

thanks for your detailed interest in our tool.

First a counter question: you are using the parameter --helixMinPu=0.7 but --model=P instead of the helix-based model --model=B... so currently your helix constraint is ignored when using the probability-focused model... are you aware of this?

Concerning your question about reasonable sortings/filtering... eventually it depends on your research question and what are you looking for and in what context..

Generally, lower energy within the model should relate to stronger binding in nature. So investigating top-ranked interactions based on E is reasonable and a standard approach.

The ensemble probability P_E provides a measure how "special" the interaction is among all other considered interactions. The latter is important since (a) the ensemble only covers interactions that heed your constraints, i.e. it is parameter dependent and (b) the ensemble computation is a heuristic in IntaRNA and not exhaustive... Thus: use P_E with care as a guide but dont put too much emphasize on the absolute values!

Nevertheless, a high P_E of the minimum free energy interaction indicates that there is no other comparable interaction alternative, which might be a good/desirable feature dependent on the studied RNAs. If there are two or more alternative interactions that show more or less the same interaction energy, the probabilities will signifacantly drop..

Unfortunately, I have no good threshold at hand for P_E.. best use it as a sorting criteria if you have a couple of "best interactions" (based on some other features) to define your experimental order etc.

Hope that helps,
best,
Martin

[MicSonn]

Dear Martin,
thank you for your detailed answers.

Regarding the --helixMinPu=0.7 parameter, I was hoping it wouldn't affect the output of the --model=P. I only realized this afterward. You get the point, thanks.

My goal is to select a given number of human genes from the output of --model=P to run a functional enrichment analysis.
At this point, I have two possible options:
a) select with E <= -26 AND the highest P_E (I'll see how to estimate a cutoff) on the output of --model=P;
b) take the intersection of --model=P AND --model=B AND --model=S with E < -26 and run functional enrichment analysis on this intersection.

Thank you very much!

Best,
MicSonn