Good Thresholds For Miranda Microrna Predictions

Question

[Doo]

Dear community,

since there are very experienced scientists among you, I would like to ask for some expertise:

Aim is to compile miRNA predictions to get a table Genes vs microRNAs. I am absolutely aware that there is a tradeoff between false positives and false negatives, but if you had to make some "yes" or "no" statements if a microRNA is targeting a gene or not what thresholds would you use for Miranda (http://www.microrna.org).

There is a SVR score and a conserved score. What would you recommend using?

Alternatively, do you know a resource where I could download those (even from multiple databases)?

Thank you very much. Help is very appreciated.

[Andrew]

Thoughts about www.microrna.org compares with the predictions from TargetScan?

Answer 1

[Larry_Parnell]

From my colleague, who has done a fair amount work in this area:

The mirSVR score provided at mircrorna.org is the best tool for making predictions, as it utilizes the most recent miRanda prediction rules such as seed-site pairing, site context, free-energy, and conservation. I would recommend a mirSVR cutoff of <= -1.2. This value represents the top 5% of miRSVR scores, where the expected probability of observing a log expression change of <=.5 is ~50%, or of <=0.1 is 70%. The authors of the SVR score also suggest conservation as merely another factor to consider when determining the overall confidence of the prediction, and not as an absolute cut off.

Also if you are using miRanda locally on your own sequences, the generally recommended values for pairing score is >150 and energy score < -7. However, others have used the more stringent values of a pairing score of >155 and an energy score of < -20. I'm hoping soon they will release an updated miranda algorithm that provides local calculation of mirSVR scores.

This reference describes how the mirSVR score was put together: Betel, D., Koppal, A., Agius, P., Sander, C. Leslie, C, et al. (2010) Comprehensive modeling of microRNA targets predicts functional non-conserved and non-canonical sites. Genome Biol, 11, R90.

Edit on 17 Oct 2011: When I wrote the above answer, we were preparing analyses that have since been published in an article entitled A Genome-wide survey for SNPs altering microRNA seed sites identifies functional candidates in GWAS. It can be accessed here and may be helpful to show some "how-to" situations.

[Nicolas Rosewick]

In miranda output, did you look at the Total Score or the maximum score for each scan ? same question for the energy

[Larry_Parnell]

There is just a score, no total or maximum score. Perhaps you used or are using an older version of the software where more options and more scores were available.