In my research, I was integrating genome-wide predictions of four microRNA target prediction algorithms (TPAs) because time of the predictions of those four TPAs were the latest and their microRNA coverage (numbers) seem to be more than that of others. However, one of the complains/comments of reviewers of my paper was why I didn't include other TPAs to my research. Therefore, I wonder how or based on what factors other researchers choose microRNA TPAs.
Hands down: Targetscan. Although all of these algorithms are limited in their usefulness, Targetscan seems to have a solid foundation and seems to predict REAL target genes, mRNAs that are actually regulated by miRNA (http://lens.elifesciences.org/05005/index.html). Combining results from different algorithms has not been very successful: there is one publication that did check the value of doing this and they came to the conclusion: don't do it (however, can't find that paper anymore).
Hands down: Targetscan. Although all of these algorithms are limited in their usefulness, Targetscan seems to have a solid foundation and seems to predict REAL target genes, mRNAs that are actually regulated by miRNA (http://lens.elifesciences.org/05005/index.html). Combining results from different algorithms has not been very successful: there is one publication that did check the value of doing this and they came to the conclusion: don't do it (however, can't find that paper anymore).
So far, I really like DIANA Tools. Have a look here: http://diana.imis.athena-innovation.gr/DianaTools/index.php
They have a number of useful applications that include exp. verified miRNA!
mirbase.org - for target prediction, it looks to contain a large number of sites and methods (above mentioned too).
miRWalk2.0 is a useful starting point as it brings together information for different sources.
However, if you're after just one source I would also recommend TargetScan. It uses a logical approach and is nicely presented.
For a paper, you should really justify why you used a particular method. The same is true for the reviewer though - they should justify why they're questioning your choice. More likely though they're just suggesting you use others for no other reason than it was an easy thing to comment on.
Targetscan is really solid, especially with their latest update which brings in functionally relevant miRNAs. However, I also like microT-CDS from DIANA Lab as the reliability has vastly improved over previous interations, it's intuitive and includes regions beyond 3' UTR.
I would like to add that although many miRNAs are predicted to target many mRNAs, actual functional miRNA-target relationships can be linked to many other factors such as expression levels, sequence conservation and base pairing. High throughput miRNA-target validation techniques are substantially contributing to the confidence of functional miRNA-target relationships. Databases such as TarBase and starBase have been hugely expanded and have become viable research tools for de novo functional relationships.
Dear Bahtiyor
Targetscan database will provide target gene based on seed complematary method. This is most efficient methods to predict target gene.
For more info.please
http://www.exiqon.com/microrna-target-prediction
I normally use the software IPA (ingenuity pathway analysis). The microRNA Target Filter in IPA uses TargetScan and also provides insights into the biological effects of microRNAs, using experimentally validated interactions from TarBase and miRecords.
Hi there
Very interesting discussion. I am not a particular fan of any method, since all of them have limitations. Some of them are very optimistic and some others more restrictive. If you are depressed, the optimistic tools are strongly recommended.
In my modest opinion, without disregarding all the elegant comments already posted, I will recommend you to use multiple predictors. The idea is to use simultaneously several algorithms for data prediction and check how many of them predict a particular target. The more algorithms predicted a target, the more credible will be. There are a few of these in the market. MirWalk2 is the golden standard since it is very well documented and updated. Also MirDIP is nice if you work with humans. Another classical tool was MiRecords, but looks that it disappeared from the web, and I cannot find an updated link
Hope it helps. All the best and thanks for the post
Paco
A brief answer: also TargetScan (including the v7 version of the eLife paper as mentioned by Thomas Andl, even if I know way better v6.2). In some cases, I suggest to keep non conserved in your list. But anyway that's a difficult question!
More insights: No TPA is perfect and they don't share the same predictions: to illustrate, we compared predictions of TargetScan v6.2 and Diana v3 (Fig S9 of Ricky et al. SciRep, 2015), and found a bimodal distribution with only around 55% and 75% of overlap among target predictions! We used these two algorithm as they use 2 different 3'UTR database of mRNA to do predictions.
As a consequence, a good compromise answer would depend on what is your objective! Tarbase is also a good option, but is far from exhaustive with a strong bias on well studied microRNA-mRNA pairs.
Last but not least, also keep in mind that predicted target mRNAs may not be transcribed at all in the cell type you're considering.
Data Bhajun Guyon Gidrol SciRep 2015 Suppl mir940 network mirna
- Badges
- Science topic