WIP: Getting cleaner and less error prone HMM parsing ready

[jamaliki]

Hey @althonos

I'm not sure you remember, but we had a chat here about getting ModelAngelo's HMM parsing all done with pyhmmer's utilities: althonos/pyhmmer#40

I think it is close right now, but it seems to be having an issue with the consensus part of the HMM. Could you take a look if you have a chance? I get the following error:

Traceback (most recent call last):
  File "/Users/kiarash/Desktop/code/cambridge/3dem_model_angelo/model-angelo/model_angelo/utils/aa_probs_to_hmm_new.py", line 93, in <module>
    hmm = aa_log_probs_to_hmm("test", probs, confidence)
  File "/Users/kiarash/Desktop/code/cambridge/3dem_model_angelo/model-angelo/model_angelo/utils/aa_probs_to_hmm_new.py", line 83, in aa_log_probs_to_hmm
    hmm.validate()
  File "pyhmmer/plan7.pyx", line 3220, in pyhmmer.plan7.HMM.validate
  File "pyhmmer/plan7.pyx", line 3246, in pyhmmer.plan7.HMM.validate
ValueError: Invalid HMM: p7H_CONS flag down, but consensus string is present

There's quite a bit of code, but what is really important is line 66 onwards. If you wish, you can see how we do it in aa_probs_to_hmm.py

[althonos]

Hi @jamaliki,

This error was caused by an outdated fix (althonos/pyhmmer#5) which was causing the consensus flag to be always set. Since the update to v0.10.0 it shouldn't be needed anymore (but I forgot to remove it), so I'l make a new patch release. Make sure to require at least PyHMMER v0.10.1 and you should be all set! :)

[jamaliki]

Thanks, worked like a charm @althonos! Going to do some testing but it should be the same. Are there any ways to write the composition row as well?

[althonos]

You don't have to set it manually, it will be computed from the HMM occupancy, just call HMM.set_composition after you're done filling it (typically right before HMM.validate). Then the consensus attribute of the HMM will be a VectorF that you're free to edit if you want.

You can also call HMM.set_consensus if you wish to build or record a consensus.

[jamaliki]

Thanks @althonos, that's great!