Variable results with cufflinks 2.0.2 using multiple threads

[david.eby]

I've been doing some testing of the cufflinks 2.0.2 tools and have been getting variable results when using multiple threads. That is, running cufflinks on the same BAM file multiple times with (for example) '-p 8' gives different results in the output files on each run. Likewise for cuffmerge (I haven't checked cuffdiff yet - that's next). I'm testing on the data from the March 2012 Nature Protocols paper using the pre-built Linux binaries.

Running single-threaded seems to be fine and always give identical results (the cuffmerge abundance results vary slightly, though that seems to be in the realm of rounding differences).

Is this a known issue? From the website it looks like there was a similar issue in v2.0.1. Are there any workarounds other than running single threaded? The multithreaded code is so much faster.

[david.eby]

To reply to my own question: after doing much more testing it looks like everything is fine. The differences that I was seeing were mostly related to a combination of ordering of data in the output files and the variable way in which the 'CUFF' identifiers are assigned.