Xchem deep

[tdudgeon]

Initial version currently bypassing GPU execution by use of the --mock flag

[simonbray]

@bgruening, what is actually the goal of this PR? Do we want to deploy the tool on usegalaxy.eu? If so we should probably remove the --mock tag and do some more testing.

[bgruening]

Yes the goal is to deploy it on usegalaxy.eu. But before we need to test it with docker on some GPU enabled host without the --mock option.

[simonbray]

But we should do that before merging here, right?

[bgruening]

this should be a tagged version

[tdudgeon]

Don't have one yet. But we will at some stage soon.

[bgruening]

no citation?

[tdudgeon]

I don't think the paper has been submitted yet.

[simonbray]

you can still write a citation in bibtex (e.g. linking to a github repo) - we also do this for rdkit because there is no publication

[tdudgeon]

I've asked if there is anything we can use.

[bgruening]

But we should do that before merging here, right?

I had hoped @tdudgeon has done this. But yes we should do the testes before :)

[bgruening]

@tdudgeon you mentioned somewhere that Docker needs to be specially enabled for GPU, can you add some infos about this?

[simonbray]

Can you reduce the size of the test data?

[tdudgeon]

Can you reduce the size of the test data?

Done tdudgeon@8b955eb

[tdudgeon]

@tdudgeon you mentioned somewhere that Docker needs to be specially enabled for GPU, can you add some infos about this?

The docker daemon needs to be a special one that supports GPUs.
I think this describes the details: https://github.com/NVIDIA/nvidia-docker
@alanbchristie knows the details. He can advise on this next week.

[bgruening]

@tdudgeon @alanbchristie have you ever tried singulairy for GPUs?

This PR is fine for me it needs some testing on a real GPU enabled host.

[tdudgeon]

have you ever tried singulairy for GPUs?
Not yet. We plan to do so.

[reskyner]

Hi Björn, can we get this merged and tested on the real infrastructure? I think there's a lot of difficulty for us getting access to a GPU to test here.

[reskyner]

Hi @bgruening @simonbray,

Have we gotten any further on this? We're currently working on a fragment screen for coronavirus. It would be awesome if we could get this tool deployed to work in sync with that screen. (to suggest new compounds)

Let me know if there's anything else we need to do

Thanks,
Rachael

[bgruening]

I was trying to run it locally, but get this ...

|  Traceback (most recent call last):
|  File "/home/bag/.planemo/planemo_tmp_rWhH6c/xchem_deep_score.py", line 204, in <module>
|  main()
|  File "/home/bag/.planemo/planemo_tmp_rWhH6c/xchem_deep_score.py", line 200, in main
|  execute(args.input, args.receptor, args.outfile, mock=args.mock)
|  File "/home/bag/.planemo/planemo_tmp_rWhH6c/xchem_deep_score.py", line 180, in execute
|  scores = read_predictions()
|  File "/home/bag/.planemo/planemo_tmp_rWhH6c/xchem_deep_score.py", line 125, in read_predictions
|  with open("{0}{1}{2}".format(work_dir, os.path.sep, predict_file_name), 'r') as input:
|  FileNotFoundError: [Errno 2] No such file or directory: './predictions.txt'

open("{0}{1}{2}".format(work_dir, os.path.sep, predict_file_name), 'r')

This is also wired? os.path.join(work_dir, predict_file_name) should work, but why r if this is an output file?

[tdudgeon]

@bgruening It runs locally for me:
planemo t --docker xchem_deep_score.xml

Possibly the docker image was not up to date (though I doubt it). I just pushed the version I used, and also tagged it as 1.0.

The predictions.txt is opened in read mode as by that stage it has been created and it's being read to extract the scores.

[simonbray]

(Please ignore this comment.)

[tdudgeon]

Damn it! My hacking about lost the mock part and symlinking the ligands.
This has now been fixed in 8641398
Remember to add the --docker arg when you run planemo!

[tdudgeon]

I don't know if it's possible to tell planemo that it needs to run with --docker and with --gpus all.
Without both we can't run proper tests with and without a GPU.