As you already mentioned there are number of prediction programs available: miRnada, microcosm, Targer Scan, Pic Tar. All programs use seed-pairing as a main criteria. PicTar allows perfect and imperfect seed complementarity. Both TargetScana nd PicTar take into account the evolutionary conservation. Target Scan adds a "context score" which improves predictions for non-conserved sites. Those miRNAs that have hight context score + and multiple binding sites have the highest chance to be true targets. miRanda allows for GU wobbles and mismatches and filters results based on the deltaG - e.g, miRanda predictes sites for let7 gene lin-41 which contains a bulge or a GU wobble in a seed region.

I personally run TargetScan+PicTar and look for the common conserved targets to select candidates for validation. Depending on how long is the list of potential targets I look in the first place at those which have multiple sites, 8 or 7mer, and are relevant to my research.

Here is a good paper on that topic: http://www.nature.com/nmeth/journal/v3/n11/full/nmeth954.html

Hi Véronique and everyone

I agree with Anna about the degree of "optimism" of some predictors. To avoid that I would suggest to use the so called "multiple" predictors, which are basically platforms that will scan for several prediction algorithms and present you the results. The more algorithms predict a particular target, the more reliable will be this prediction.

I have compiled some of them in my Mirnatools page.

https://sites.google.com/site/mirnatools2/multiple-predictors

Hope it helps. Best luck

Paco

Thank you very much to everyone for your prompt answers. I will look at these algorithms in combination.

I really appreciate your help,

Regards,

Veronique