Prediction of microRNA targets

I have a set of mRNAs AND a set of microRNAs. I want to obtain (potentially) interacting microRNA-mRNA pairs from those sets. I’m working with an organism with too few experimentally validated interactions, so the task is a bioinformatical prediction.

This “target prediction” goal can look pretty simple, but to me it isn’t: there are too many ways to do this. There are Targetscan with its metrics based on conservative sites; blast-like Miranda; mirDB with its own MirTarget algorithm; RNAhybrid and more. And I figured out that the results of these tools do not overlap well.

Now I am using Miranda and RNAHybrid, both with strict cut-off to reduce the number of false positives (but there are still a lot of them), however, it doesn’t seem to be an optimal approach. So: are there any “good practices” to bioinformatically predict microRNA-mRNA interactions? What approaches have you tested or established in your lab?

ncrna targetscan microrna mirdb miranda

• 1,943 views • link

written 2.3 years ago by ivmstalker •

1 answer

Sometimes I feel like certain problems in bioinformatics compare to prescription-free remedies for the common cold, a lot are on offer but the sheer amount indicates none of them really works well. A common approach could be to try several remedies in combination and look if there is any consensus. I found this review which seems like a good read: https://doi.org/10.3390%2Fmps4010001

You are already combining two. A pragmatic approach may be to test 3-4 different methods (there is no "best" so you need to make your choice by some of the criteria in the review). Then be prepared for any additional method the reviewers favor in addition.

The main question is what is the intention behind analyzing this organism, do you want to publish the data, and are you therefore prepared to experimentally validate the results. If you are dealing with a non-model organism where little work on miRome has been done you could be fine. Se f.e. this article https://link.springer.com/article/10.1007/s10126-021-10058-z#Sec2 on a sea louse which used the same tools you proposed and got by with qPCR validation alone. So if you are by chance working with similarly weird organisms, chances are you can publish your findings later on.

• 0 views • link

written 2.3 years ago by Michael Dondrup

Yep, caught this cold. My research won't die, of course, but it will have to stay in bed for a few days

More tools is a really obvious solution, I'll use it if my research runs out of sick leave. Seen articles where the authors just used more tools so they don't have to change the default settings of each tool :)

Still looking for a less crude solution while there's still time, though.

Experimental validation of some pairs will be conducted. But we have a large number of differentially expressed molecules, and I hope to include miRNA-mRNA data for all of them in my research. With all necessary limitations stated, ofk - don't want someone be as frustrated as I was when I discovered how inaccurate these predictions are.

Thanks for the reply and for the links! I like this article. Testing and comparing tools. With the recommended parameters from this paper Miranda and RNAHybrid still predict too many false positives, but it might be a good start to... Well, let's call it "empirical parameter selection".

• 0 views • link

written 2.3 years ago by ivmstalker •