yarden
commented
10 years ago Hi,
Could you please send me a BAM file and a few events where this happens? Ideally a small BAM file with just the events that show this. Also, let me know what MISO version you used and how you ran it.
A few general thoughts:
- We had a bug a while ago that caused instability in the Bayes factor calculation, making it vary potentially 10-fold (on low coverage events) across runs. That means that in some run, the BF might be 5 and on another run it would be 50, so depending on the cut off, the event would or wouldn't be in your 'filtered' set. I believe this was fixed, but it's possible that there's an edge case I missed.
- MISO by default applies essentially no read coverage filter. It requires only a small number of reads in the entire locus/event. So if you have an alternative exon with 1 read in it and a bunch of scattered reads in flanking exon, it will still produce Psi values, and those will mostly be noise. It's up to the user to filter on the counts to make sure there are a sizeable number of reads supporting inclusion or exclusion across the sample set. It's possible that the event that you see variability in are those, in which case the solution would be to apply a read counts filter.
It's hard to say without an example. Please send a link to the files either to me by email, or you can also post them on here.
If it is a bug on our end, I'd be very keen to fix it ASAP, but hard to do without a specific example that you've run into.
Thanks. Best, --Yarden
Hi, I'm using Miso to estimate exon skipping alternative events. When I repeat the analysis several times (using the same files) I have the same events in the comparisons output file with the bayes factor and PSI values that change every time. The result is that the filter.py command gives me every time a different set of alternative events. In your opinion, whtat's the mistake? Maybe should I set some 'seed' parameters in the settings file?
thank you