I would like to know if anyone has any recommendation on how to filter out false positive junctions detected from RNA-seq data. I am currently using different tools to detect splice junctions, but there are thousands of junctions that are detected. One way to filter out false positive junctions would be to set a threshold on how many reads support the junction and/or in how many samples the junction is found. However, it is difficult to set a common threshold when samples have different coverage, and some junctions with a low number of supporting reads and/or a low number of reoccurrence can still be of interest.
What kind of approach do you advise using? I know that some tools provide linear regression models to filter out FPs.