Understanding miRGate scoring
Issue #86
resolved
Hi, I'm currently using miARma-seq to analyse miRNA data. I ran the TargetPrediction module of miARma-seq and got some excel files. MiRGate use many mRNA prediction methods. Does the scoring in miRGate be the score of the method which is used? For example, if the Miranda method give a score of 1.25. It means that the scoring of the interaction is based on Miranda algorithm?
Do you have maybe some score threshold for Miranda, Tagertscan and Pita?
Thank you
Sébastien
Comments (9)
-
-
reporter
That's a great idea I think! Concerning the score, do you have some score threshold that you can advise me?
Sebastien
-
Difficult to say. All predictions stored in miRGate are over the threshold of each method. For example miRanda recommends predictions with a score greater that 120 (it should be 1 in the normalised score you mention earlier), so we only provided interactions > 120 for miranda. The same for each predictor. So, mainly all predictions should be good. But in my opinion (more than 10 collaborations including miRNA-mRNA validations) the agreement value is the best way. If no agreement value is found, we manage to validate several interactions with a normalised score >=1.5
Edu
-
reporter
By agreement you mean that the same miRNA-mRNA interaction is found by the tools at least 10 times?
Thank you for your help ;)
Sebastien
-
Nop, agreement value is when a prediction is predicted by more than one predictor in the same genomic coordinate.
As an example: http://mirgate.bioinfo.cnio.es/ResT/API/human/miRNA_gene_predictions/hsa-mir-21-5p/PDCD4
If you take a look to these 3 predictions , the first one (id: 158159178) has an agreement value=1. Nevertheless, the id 158159177 has a value=2. If you see all values, you will see than among the coordinates 228-249 of the ENST00000280154 3'UTR, miRanda and Targetscan provide a common prediction.
-
reporter
Ho I see, thank you. Is there a way to get the agreement values from the xls files? Or must I process the file to get it?
-
unfortunately at this moment it is not possible. But in a month or so, It will be included by default
-
reporter
ok thank you ;) i'll use a normalised score of at leat 1.5 for this time :)
Sebastien
-
reporter
- changed status to resolved
- Log in to comment
Hi! Nop, this score is a normalised score among all predictors (It's the only way to compare predictions). I'm working in a new miARma version, which will include more information in this section. My idea is to include the raw score for each predictor and the agreement value (Number of different predictor that support a prediction within the same genomic coordinates)
Edu