No bonds beween atoms...?

[spyda90]

Hi,

Thank you for developing NeuralPlexer. We ran our first attempts on a very challenging protein and we obtained promising results, most of all very different from the (useless) ones obtained with our tools.

However, the ligand protein complexes we obtained have a "normal" protein for some amino acids streches while in other ones many bonds are missing (for example, a Leucine residue is represented with all the atoms (putatively at the right distances) but no bonds are there).
The longest stretch where this happens is about 50aa.

In order to proceed with other studies (e.g. MDs) we'd like to solve this issue. What can be the reason why is this happening?

Thank you in advance.

Francesco

[Luke-ebbis]

Hey @spyda90 is this issue related to #40 ? Can you share more information, I'd like to reproduce it

[spyda90]

Hi,

It may be similar to #40 but it happens in many other atoms.
Here you are the command line:

neuralplexer-inference --task=batched_structure_sampling --input-receptor 'SHMELTPDQQTLLHFIMDSYNKQRMPQEITNKILKEEFSAEENFLILTEMATNHVQVLVEFTKKLPGFQTLDHEDQIALLKGSAVEAMFLRSAEIFNKKLPSGHSDLLEERIRNSGISDEYITPMFSFYKSIGELKMTQEEYALLTAIVILSPDRQYIKDREAVEKLQEPLLDVLQKLCKIHQPENPQHFACLLGRLTELRTFNHHHAEMLMSWRVNDHKFTPLLCEIWDVQ|KENALLRYLLDKD' --input-ligand FXR/ZGUG.sdf --out-path FXR/ --model-checkpoint neuralplexermodels_downstream_datasets_predictions/models/complex_structure_prediction.ckpt --n-samples 16 --chunk-size 4 --num-steps=40 --cuda --sampler=langevin_simulated_annealing

This is how it appears:

Carbons and some nitrogens are there. No oxygens, no bonds. Where ribbons are present, amino acids are fine.

Please let me know if you need something more.

[spyda90]

Hi, any news about this issue?

[Luke-ebbis]

Hey Spyda, I have no news on the issue, I've been waiting too. If you send me your SDF file, I can try to run it on my system. Then we can see whether the problem persists on different platforms.

[spyda90]

Hi, how can I send you a SDF file? The system do not let me send SDF here. Thank you

[Luke-ebbis]

Hey @spyda90, you could deposit it on zenodo, or specify a download link (like I did in #49). Looking forward to seeing what my installation does.

[Luke-ebbis]

@spyda90 I found this structure

See the attachment

issue.zip

Call was

apptainer run --nv results/dependencies/neuralplexer/neuralplexer.sif neuralplexer-inference --task=batched_structure_sampling --model-checkpoint results/dependencies/neuralplexer/data/neuralplexermodels_downstream_datasets_predictions/models/complex_structure_prediction.ckpt --out-path results/data/issue-41/issue-41/ --input-receptor 'SHMELTPDQQTLLHFIMDSYNKQRMPQEITNKILKEEFSAEENFLILTEMATNHVQVLVEFTKKLPGFQTLDHEDQIALLKGSAVEAMFLRSAEIFNKKLPSGHSDLLEERIRNSGISDEYITPMFSFYKSIGELKMTQEEYALLTAIVILSPDRQYIKDREAVEKLQEPLLDVLQKLCKIHQPENPQHFACLLGRLTELRTFNHHHAEMLMSWRVNDHKFTPLLCEIWDVQ|KENALLRYLLDKD' --input-ligand 'data/ZGUG.sdf' --sampler langevin_simulated_annealing --n-samples 1 --chunk-size 1 --num-steps=5 --cuda

[spyda90]

Hi,
In Pymol everything seems fine, but loading up in Maestro you can see how there are no connections between several atoms (still).

So, unfortunately, the issue is still there...