Hi All, I have a question regarding parallelization in Mutect2. If I understand correctly, one can provide each Mutect2 run with a BED interval file, and you can use a list of intervals to run multiple instances on multiple threads. As far as I see, this parallelization on Mutect2 is not implemented on Sarek. Is there any technical obstacle or consideration to implement it?

Mutect2 can run with the `--intervals` parameter (assuming that is what you mean). With WES, this is (in my experience) not being used for parallelization (may not be efficient), since that BED contains all intervals. But I think it is being used to parallelize analysis of WGS samples, since that has many different BED files.

Thanks for explanation. However, it is my understanding that this feature is not being used efficiently. The option to use intervals is supposed to facilitate multiple instances simultaneously, each running on one interval. However, the way the process is working now there is no parallelization, and the intervals are being fed to Mutect2 or FreeBayes in a serial manner. Next interval is not being processed until the process of current interval is complete. That is why they take so long. Am I understanding it correctly?

For WGS I see multiple intervals files being used, each in a separate process (I think, I don't have the work dir any more), and those can run in parallel. (Of course any merging has to wait for all to complete.) In WES, where the intervals are short, I just see one BED file being used to restrict the calling space to the targets. (I assume those could be split, but splitting & merging has its own costs, which may exceed the benefits. I think the "nucleotides per second" parameter controls that trade-off.) @Maxime Garcia, @Rike Hanssen, do I have this right or is there something to add?