salmon 0.4.0 core dump for genes

[Ido Tamir]

Hi,
I use salmon 0.4.0 and I get a core dump for the genes. It worked with 0.3.0 for the same input and parameters.
It still works for the transcripts (including bias estimation). But before it creates the genes file it fails.

best,
ido

[2015-06-08 20:17:20.458] [jointLog] [info] iteration = 6300 | max rel diff. = 0.0109258
[2015-06-08 20:17:22.118] [jointLog] [info] iteration = 6400 | max rel diff. = 0.0105151
[2015-06-08 20:17:23.777] [jointLog] [info] iteration = 6500 | max rel diff. = 0.0103964
[2015-06-08 20:17:25.442] [jointLog] [info] iteration = 6600 | max rel diff. = 0.0103901
[2015-06-08 20:17:27.108] [jointLog] [info] iteration = 6700 | max rel diff. = 0.0103723
[2015-06-08 20:17:28.765] [jointLog] [info] iteration = 6800 | max rel diff. = 0.0103228
[2015-06-08 20:17:30.340] [jointLog] [info] iteration = 6900 | max rel diff. = 0.0101859
[2015-06-08 20:17:31.293] [jointLog] [info] iteration = 6964 | max rel diff. = 0.00999706
[2015-06-08 20:17:31.299] [jointLog] [info] Finished optimizer
[2015-06-08 20:17:31.299] [jointLog] [info] writing output

parsed 102311 features
parsed 1 expression values
Performing PCA decomposition
salmon: /home/vagrant/salmon/include/eigen3/Eigen/src/Core/Redux.h:202: static Eigen::internal::redux_impl<Func, Derived, 3, 0>::Scalar Eigen::internal::redux_impl<Func, Derived, 3, 0>::run(const Derived&, const Func&) [with Func = Eigen::internal::scalar_sum_op<double>; Derived = Eigen::Block<const Eigen::Matrix<double, -1, -1>, -1, 1, true>; Eigen::internal::redux_impl<Func, Derived, 3, 0>::Scalar = double]: Assertion `size && "you are using an empty matrix"' failed.
/groups/csf-ngs/bin/align/SalmonBeta-0.4.0_DebianSqueeze/salmon.sh: line 7: 31160 Aborted                 (core dumped) /groups/csf-ngs/bin/align/SalmonBeta-0.4.0_DebianSqueeze/bin/salmon "$@"

[Rob]

Hi Ido,

  Thanks for reporting this.  The issue is a result of the fact that the output format has changed in version 0.4.0 (we no longer report FPKM), but the post-hoc bias correction code wasn't updated to account for this. I've fixed this and uploaded an updated binary to the releases page.  If this fixes the issue for you, I'll bump this to minor release.  I must say that we're moving away from the post-hoc bias correction methodology (which is part of the reason this bug popped up in the first place).  Most types of bias are now accounted for during the inference itself.  Once sequence-specific bias is accounted for during the inference, I'll likely remove the post-hoc bias correction option, as it will be redundant.  Thanks again for reporting this, and let me know if it solves your problem.

--Rob

P.S. I've also turned off the "annoying" reporting of each fragment that maps in an un-expected orientation.