fold_analysis_latest.sif generation scores problem

[Qrouger]

Hi, the latest version of the Singularity image for generate scores don't calculate pi_score for me.

predictions_with_good_interpae.csv

[DimaMolod]

yes, this is because we removed ccp4 from the image. Please refer to #472

[Qrouger]

I'm sorry but I cannot generate the pi_score by following the image build with ccp4. It's maybe cause by the ccp4 version (v9.0.005) ?

[DimaMolod]

I am sorry to hear that. Yes, it could be because of incompatible ccp4 version, could you try to downgrade ccp4? Can you also provide log files with exact exceptions? If they are not informative, there is a trick you could try:
run your singularity image interactively with this command:
singularity shell /path/to/container.sif
...and try to run pi score script from within the container using these scripts:
https://github.com/KosinskiLab/AlphaPulldown/blob/main/alphapulldown/analysis_pipeline/run_pi_score.sh
https://github.com/KosinskiLab/AlphaPulldown/blob/main/alphapulldown/analysis_pipeline/run_get_good_pae.sh
the chances are that this way you will be able to get a more comprehensive error message.

[Qrouger]

Hi, I can't found solution to downgrade ccp4.
I run singularity exec --no-home --bind /data/result_all_vs_all:/mnt ~/Downloads/ccp4-9/fold_analysis_latest_withCCP4.sif run_get_good_pae.sh --output_dir=/mnt --cutoff=10

Log files don't show problems.

log_fold_analysis.txt

I don't know when run_pi_score.py/.sh are used.

[jstigler]

FWIW, the pi_score calculation script appears commented out in run_and_summarise_pi_score:

 # We decided to remove pi_score from the pipeline
                #result = subprocess.run(command, capture_output=True, text=True)
                #if result.returncode != 0:
                #    return self._handle_pi_score_error(result, command, result.stderr)

I still end up with errors though:

All features could not be calculated successfully for atleast one of the interfaces!\n
Exiting\n
stderr=' SC: mds: too many atoms\n')
Command: ['bash', '-c', 'source activate pi_score && export PYTHONPATH=/software:$PYTHONPATH && python /software/pi_score/run_piscore_wc.py -p /mnt/XXXX_and_YYYY/ranked_0.pdb -o /tmp/tmphidm1__z -s 2']. Error: SC: mds: too many atoms

[jkosinski]

Yeah, we commented it out as a quick fix after removing CCP4 from the default installation, assuming that pi_score might not be essential. @DimaMolod, maybe we could add a conditional check for CCP4 availability and run it if present?

The "Error: SC: mds: too many atoms" message looks familiar. We encountered it occasionally with large or highly clashing structures. It originates from the SC program in CCP4 and has not been resolved so far. Does the analysis pipeline crash entirely, or does it just skip this model?

[DimaMolod]

Yeah, we commented it out as a quick fix after removing CCP4 from the default installation, assuming that pi_score might not be essential. @DimaMolod, maybe we could add a conditional check for CCP4 availability and run it if present?

yes, we removed pi_score because it doesn't work with the newer CCP4 version, and AFAIK there is no way to install the older version...the underlying reason is that pi_score is written in python2.7

The "Error: SC: mds: too many atoms" message looks familiar. We encountered it occasionally with large or highly clashing structures. It originates from the SC program in CCP4 and has not been resolved so far. Does the analysis pipeline crash entirely, or does it just skip this model?

I am afraid this error message is generic and can mean a variety of actual problems.
@jkosinski from the feedback I collected so far, this whole approach with re-installing CCP4 locally is very unstable and needs to be revisited ASAP.

[jstigler]

Does the analysis pipeline crash entirely, or does it just skip this model?
It keeps going with the other models. It just doesn't fill the corresponding entries in the table.