[miso-users] bayes factor calculation not consistent among replicates

[Yarden Katz]

Hi James,

See replies below:


On Jan 27, 2014, at 11:52 AM, Gao, James (NIH/NCI) [E] wrote:

> Hi Yarden,
> 
> I am comparing alternative splicing events in a pair of control/treatment
> samples. Each has two replicates. If I looked at the junction reads in
> UCSC genome browser, the two replicates look similar and one exon is
> clearly spliced differently between control and treatment. However, only
> replicate #1 comparison generated a bayes factor, not replicate #2. I
> looked at the miso calculation, it seems there are no .miso files created
> for the 2nd replicates for that particular exon junction. Not quite sure
> what controls the generation of .miso files.


If there's no .miso file, that means MISO either failed to process that event (for technical reasons) or it failed to meet your filter criteria.  A technical reason might be that you ran the jobs on a cluster (chunking the events into bins) and that some chunk, for example, failed or got killed in middle.  A non-technical reason might be that the event fails to meet the read coverage criteria that is set in the MISO configuration file.  If at least N reads are required at a locus, it could be that replicate 2 fails to meet those.  You can easily test this by looking at what reads are in replicate 2 in that event using samtools, or you could run MISO with a GFF containing just that event on replicate 2.

> 
> Another non-related question about sashimi-plot, how the # of exon
> junction reads are decided? They seem to be different from the
> junctions.bed file that TopHat generated.

To give a definitive answer, it'd be helpful if you give an example of a single GFF event and a BAM file with just reads for that event where the two give different answers.  Note that Sashimi plot skips reads with insertions/deletions, which Tophat might not depending on the settings you ran it with (and whether with Bowtie 1 or Bowtie 2).  

Best, --Yarden

> 
> Thank you.
> 
> James
> 
> 
> _______________________________________________
> miso-users mailing list
> miso-users at mit.edu
> http://mailman.mit.edu/mailman/listinfo/miso-users