Salmon Crash

[Gavin Wilson]

Hello,

I have been using salmon successfully on 31 out of my 32 samples, however, I get a crash every time I process one of them.

command:

salmon quant -i ./transcripts_index -l IU -1 r1.fastq.gz -2 r2.fastq.gz -p 14 --useVBOpt --biasCorrect --useFSPD --output ./salmon/

Output log:

Version Info: This is the most recent version of Salmon.
# salmon (mapping-based) v0.6.0
# [ program ] => salmon
# [ command ] => quant
# [ index ] => { ./transcripts_index }
# [ libType ] => { IU }
# [ mates1 ] => { r1.fastq.gz }
# [ mates2 ] => { r2.fastq.gz }
# [ threads ] => { 14 }
# [ useVBOpt ] => { }
# [ biasCorrect ] => { }
# [ useFSPD ] => { }
# [ output ] => { ./salmon/ }
Logs will be written to ./salmon/logs
[2016-03-08 13:03:29.022] [jointLog] [info] parsing read library format
there is 1 lib
Loading 32-bit quasi index[2016-03-08 13:03:29.101] [jointLog] [info] Loading Quasi index
[2016-03-08 13:03:29.102] [stderrLog] [info] Loading Suffix Array
[2016-03-08 13:03:29.103] [stderrLog] [info] Loading Position Hash
[2016-03-08 13:03:30.500] [stderrLog] [info] Loading Transcript Info
[2016-03-08 13:03:30.937] [stderrLog] [info] Loading Rank-Select Bit Array
[2016-03-08 13:03:31.116] [stderrLog] [info] There were 291847 set bits in the bit array
[2016-03-08 13:03:31.149] [stderrLog] [info] Computing transcript lengths
[2016-03-08 13:03:31.150] [stderrLog] [info] Waiting to finish loading hash
Index contained 291847 targets
[2016-03-08 13:03:36.373] [stderrLog] [info] Done loading index
[2016-03-08 13:03:36.373] [jointLog] [info] done

processed 60000000 fragments
hits: 436363199, hits per frag:  7.27636d

[2016-03-08 13:09:25.666] [jointLog] [info] Computed 621501 rich equivalence classes for further processing
[2016-03-08 13:09:25.666] [jointLog] [info] Counted 59455663 total reads in the equivalence classes

[2016-03-08 13:09:25.756] [jointLog] [info] Mapping rate = 98.4492%

[2016-03-08 13:09:25.756] [jointLog] [info] finished quantifyLibrary()
[2016-03-08 13:09:25.756] [jointLog] [info] Starting optimizer
[2016-03-08 13:09:26.163] [jointLog] [info] Marked 0 weighted equivalence classes as degenerate
[2016-03-08 13:09:26.202] [jointLog] [info] iteration = 0 | max rel diff. = 46.4077
[2016-03-08 13:09:28.030] [jointLog] [info] iteration 50, recomputing effective lengths
[2016-03-08 13:09:43.253] [jointLog] [info] iteration = 100 | max rel diff. = 13.3386
[2016-03-08 13:09:46.941] [jointLog] [info] iteration = 200 | max rel diff. = 13.2108
[2016-03-08 13:09:50.636] [jointLog] [info] iteration = 300 | max rel diff. = 13.2143
[2016-03-08 13:09:54.330] [jointLog] [info] iteration = 400 | max rel diff. = 12.1461
[2016-03-08 13:09:57.987] [jointLog] [info] iteration 500, recomputing effective lengths
[2016-03-08 13:10:11.386] [jointLog] [info] iteration = 500 | max rel diff. = 34.8084
Exception : [Error in function boost::math::digamma<long double>(long double): Evaluation of function at pole -nan]
/u/gwilson/local/bin/salmon quant was invoked improperly.
For usage information, try /u/gwilson/local/bin/salmon quant --help
Exiting.

I am assuming it is the bias correction option but it would be ideal if I could use this option for all of my samples. 

Thanks.

[Rob]

Hi Gavin,

  Actually, from the message corresponding to the exception, I'm almost certain it's being caused by the `--useVBOpt` option.  It looks like the digamma function (which is used in the Variational Bayesian EM algorithm to update abundances) is encountering an invalid input.  Unfortunately, I've not encountered this error before, but I would test to see if the sample completes successfully if you leave off the `--useVBOpt` (this will optimize abundances using the standard expectation maximization algorithm, which should still be very accurate).  If this is, indeed, the problem, I'd be happy to take a look at the data to see if I can reproduce the error (if you're able to share it).

--Rob

[Gavin Wilson]

Hi Rob,

Thank you so much for the fast response.  You were indeed correct and removing the --useVBOpt option mitigated the error.  Unfortunately, I cannot share the data but I am happy to compile salmon with debugging options and get you a back trace or any other useful information.

Thanks again,

Gavin

On Tuesday, March 8, 2016 at 7:12:03 PM UTC-5, Gavin Wilson wrote:

...