GNN model refinement problem _ SBGrid and cluster

[BaudaE]

Good afternoon,

Very genuine question. I would like to run ModelAngelo through SBGrid on a cluster. I use this command : salloc -p gpu -n 1 -c 1 --gres=gpu:1 srun model_angelo build -v EMmap.mrc -f sequence.fasta -o ModelAngelo_test1

The first step Initial C-alpha prediction is fast and easy, but the program seems to run endlessly without any progress during the first round of GNN model refinement.

The log file does not give much information and finishes with : 2023-01-20 at 15:29:50 | INFO | Model prediction done, took 511.72 seconds for 512 sliding windows 2023-01-20 at 15:29:50 | INFO | Average time is 999.458 ms 2023-01-20 at 15:29:50 | INFO | Starting Cα grid to points... 2023-01-20 at 15:29:51 | INFO | Have 322 Cα points before pruning and 154 after pruning 2023-01-20 at 15:29:51 | INFO | Finished inference! 2023-01-20 at 15:29:51 | INFO | GNN model refinement round 1 with args: {'num_rounds': 3, 'crop_length': 200, 'repeat_per_residue': 3, 'esm_model': 'esm1b_t33_650M_UR50S', 'aggressive_pruning': True, 'seq_attention_batch_size': 200, 'map': 'PATH/EMmap.mrc', 'fasta':'sequence.fasta', 'struct': 'PATH/see_alpha_output/see_alpha_output_ca.cif', 'output_dir': 'PATH/gnn_output_round_1', 'model_dir': '/programs/share/modelangelo/0.2.2/hub/checkpoints/model_angelo/original/gnn', 'device': 'cuda:0'} 2023-01-20 at 15:30:25 | INFO | Loaded module from step: 301201

Do you have a suggestion to understand why the refinement does not work? Thank you very much

[jamaliki]

Hi,

No this should not happen. It does not seem like there is any issue with your fasta file. I am not too familiar with how SBGrid is running things, but could it be that the -c 1 option is only giving it one CPU?

@jasonkey do you have any ideas?

[ezbie]

Hello, did you find a solution to this? When I use build with a FASTA file the process stops running. It runs when the fasta is not supplied.

[jamaliki]

Hi @ezbie ,

Is this happening for you? Can you check to see if you are running the newest version of the software?

Best, Kiarash.