As miRNA discovery continues to skyrocket, the obvious question is: What are these miRNAs targeting? Lately, that topic seems to be as hot as any “Jon & Kate Plus 8” gossip. Several programs have been created to get at that question, and predict targets for miRNAs, but it’s hard to know which ones are really worth your time.
A group from the Biomedical Sciences Research Center in Varkiza, Greece took it upon themselves to test out popular miRNA target prediction algorithms, and publish what they found.
Using a set of known gene targets for 5 miRNAs as a benchmark, they assessed how each of these programs measured up.
- DIANA-microT 3.0
- ElMMo
- miRanda
- mirbase
- Pictar
- PITA
- RNA22
- TargetScan 5.0
- Simple seed measure
What they saw was that each of these programs have their own pluses and minuses, but all of them failed to ID a big chunk of gene targets, finding that most had a precision (correct predicted/total predicted) of ~50% and a sensitivity (correct predicted/ total correct) ranging from 6-12% when compared to their control data set. They even tried using several programs in combination, and still left a large percentage of targeted genes on the table.
The take home message here is: miRNA target prediction still has a way to go before you can ditch those validation experiments, but as knowledge of miRNAs increase it will continue to improve.
Get the complete rundown for all these algorithms at Bioinformatics, September 2009