UnboundLocalError: local variable 'dispatch_tuple' referenced before assignment

[Lcarey]

Hi,
ISS looks interesting. I'd like to simulate error rates from aligned reads.
I installed via pip3 (python 3.6.4) and tried to generate a model, but got an error:

~/.local/bin/iss model --bam test2.bam --output test
INFO:iss.app:Starting iss model
INFO:iss.app:Using KDE ErrorModel
INFO:iss.bam:Reading bam file: test2.bam
Traceback (most recent call last):
  File "/home/lcarey/.local/bin/iss", line 11, in <module>
    sys.exit(main())
  File "/home/lcarey/.local/lib/python3.6/site-packages/iss/app.py", line 417, in main
    args.func(args)
  File "/home/lcarey/.local/lib/python3.6/site-packages/iss/app.py", line 214, in model_from_bam
    bam.to_model(args.bam, args.output)
  File "/home/lcarey/.local/lib/python3.6/site-packages/iss/bam.py", line 180, in to_model
    for pos, indel in modeller.dispatch_indels(read):
  File "/home/lcarey/.local/lib/python3.6/site-packages/iss/modeller.py", line 330, in dispatch_indels
    yield dispatch_tuple
UnboundLocalError: local variable 'dispatch_tuple' referenced before assignment

I've uploaded a bam file that fails, and a subset of that file that works:
https://www.dropbox.com/sh/vl1zhz59u8nyela/AAAzAAn4fgdmzixe9cQDCa4Sa?dl=0

[HadrienG]

Hi!

Thank you for reporting this and providing test files. It also fails on my machine.

What fails is the CIGAR parsing. InSilicoSeq only handles matches, insertions and deletions at the moment. I've omitted the special (and rarer!) cases of skipped region, padding and soft/hard clipping in the parsing logic, which I think causes this error.

I'll fix this in the coming days, and come back to you with a minor version bump.

Best,
Hadrien

[Lcarey]

Hi Hadrien,

Cool.

Did you think about learning context-dependent error profiles from the .bam file? My understanding of Illumina errors is that they also depend on the -1nt, and to a lesser extent on the -2nt. (eg https://www.ncbi.nlm.nih.gov/pubmed/18660515 and others).

We're interested in using ultra-high coverage (~milllion x) coverage to identify subclonal mutations.

-Lucas

[HadrienG]

Hi!

Sorry for the delay. The bug you reported is solved with 1.2.1.
For answering your question, we're thinking to give HMMs a try, but I'm still struggling with how to best handle the dependance on the -2nt event. But hopefully there will be a HMM error profile option for version 2 😃