hannesstark / flowsite Goto Github PK

Hi Hannes!

Your work is exciting and we want to follow up.
But the link of the processed data (pdbbind-zenodo) seems to be broken recently.

Could you please help to fix it?

Looking forward to your reply.

Amazing Job Hannes!

I tried to test the code with the linked packages installed via "pip", but there are some issues on some options:

1. Activating the -wandb option, it tells that wandb has to be inizialized first (I thought that it may expect a training step before inference from provided weights)

File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/runpy.py", line 196, in _run_module_as_main
return _run_code(code, main_globals, None,
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/runpy.py", line 86, in _run_code
exec(code, run_globals)
File "/home/marco/Scrivania/FlowSite/inference.py", line 75, in
main_function()
File "/home/marco/Scrivania/FlowSite/inference.py", line 69, in main_function
if args.wandb: wandb.log({'numel': numel, 'trainer/global_step': 0})
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/site-packages/wandb/sdk/lib/preinit.py", line 36, in preinit_wrapper
raise wandb.Error(f"You must call wandb.init() before {name}()")
wandb.errors.Error: You must call wandb.init() before wandb.log()

2. The options --ns 48 and --nv 12 raised many model shape errors (i.e., "copying a param with shape torch.Size([128, 96]) from checkpoint, the shape in current model is torch.Size([128, 144])"). Using --ns 32 and --nv 8, as in the training script, solved the problem

3. The CUDA GPU is detected, but the following torch problem was raised:

File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/runpy.py", line 196, in _run_module_as_main
return _run_code(code, main_globals, None,
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/runpy.py", line 86, in _run_code
exec(code, run_globals)
File "/home/marco/Scrivania/FlowSite/inference.py", line 76, in
main_function()
File "/home/marco/Scrivania/FlowSite/inference.py", line 73, in main_function
trainer.predict(model=model_module, dataloaders=predict_loader, ckpt_path=args.checkpoint)
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/site-packages/lightning/pytorch/trainer/trainer.py", line 865, in predict
return call._call_and_handle_interrupt(
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/site-packages/lightning/pytorch/trainer/call.py", line 44, in _call_and_handle_interrupt
return trainer_fn(*args, **kwargs)
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/site-packages/lightning/pytorch/trainer/trainer.py", line 904, in _predict_impl
results = self._run(model, ckpt_path=ckpt_path)
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/site-packages/lightning/pytorch/trainer/trainer.py", line 990, in _run
results = self._run_stage()
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/site-packages/lightning/pytorch/trainer/trainer.py", line 1031, in _run_stage
return self.predict_loop.run()
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/site-packages/lightning/pytorch/loops/utilities.py", line 181, in _decorator
return loop_run(self, *args, **kwargs)
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/site-packages/lightning/pytorch/loops/prediction_loop.py", line 122, in run
self._predict_step(batch, batch_idx, dataloader_idx, dataloader_iter)
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/site-packages/lightning/pytorch/loops/prediction_loop.py", line 250, in _predict_step
predictions = call._call_strategy_hook(trainer, "predict_step", *step_args)
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/site-packages/lightning/pytorch/trainer/call.py", line 309, in _call_strategy_hook
output = fn(*args, **kwargs)
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/site-packages/lightning/pytorch/strategies/strategy.py", line 429, in predict_step
return self.lightning_module.predict_step(*args, **kwargs)
File "/home/marco/Scrivania/FlowSite/lightning_modules/flowsite_module.py", line 510, in predict_step
designed_seqs.append([np.array(RESTYPES)[designed_seq.cpu().numpy()][torch.where(full_prot_bid == i)] for i in range(len(batch.pdb_id))])
File "/home/marco/Scrivania/FlowSite/lightning_modules/flowsite_module.py", line 510, in
designed_seqs.append([np.array(RESTYPES)[designed_seq.cpu().numpy()][torch.where(full_prot_bid == i)] for i in range(len(batch.pdb_id))])
File "/home/marco/miniconda3/envs/flowsite/lib/python3.10/site-packages/torch/_tensor.py", line 1030, in array
return self.numpy()
TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first

The CPU version of torch works fine

Hi,

First of all, thanks for the code, it is very cool!

I was tyring to run the examples you provide in the repo:

CUDA_VISIBLE_DEVICES="0" python -m inference --num_inference 10 --out_dir data/inference_out --csv_file data/inference_csv_example.csv --batch_size 16 --checkpoint pocket_gen/lf5t55w4/checkpoints/best.ckpt --run_test --run_name inference1 --wandb  --layer_norm --fake_constant_dur 100000 --fake_decay_dur 10 --fake_ratio_start 0.2 --fake_ratio_end 0.2 --residue_loss_weight 0.2 --use_tfn --use_inv --time_condition_tfn --correct_time_condition --time_condition_inv --time_condition_repeat --flow_matching --flow_matching_sigma 0.5 --prior_scale 1 --num_angle_pred 5 --ns 48 --nv 12 --batch_norm --self_condition_x --self_condition_inv --no_tfn_self_condition_inv --self_fancy_init

And:

CUDA_VISIBLE_DEVICES="0" python -m inference --num_inference 10 --out_dir data/inference_out2 --design_residues "A60-65,A232,A233,A212-215,A325" --ligand data/2fc2_HEM_HBI_HAR_NO.mol2 --protein data/2fc2_unit1_protein.pdb --batch_size 16 --pocket_def_ligand data/2fc2_HEM_HBI_HAR_NO.mol2 --checkpoint pocket_gen/lf5t55w4/checkpoints/best.ckpt --run_test --run_name inference1 --wandb  --layer_norm --fake_constant_dur 100000 --fake_decay_dur 10 --fake_ratio_start 0.2 --fake_ratio_end 0.2 --residue_loss_weight 0.2 --use_tfn --use_inv --time_condition_tfn --correct_time_condition --time_condition_inv --time_condition_repeat --flow_matching --flow_matching_sigma 0.5 --prior_scale 1 --num_angle_pred 5 --ns 48 --nv 12 --batch_norm --self_condition_x --self_condition_inv --no_tfn_self_condition_inv --self_fancy_init

However, both error with the following message:

  File "/home/michaelcarter/DD_tools/FlowSite/datasets/inference_dataset.py", line 106, in get
    assert self.args.pocket_residue_cutoff is not None, 'distance pocket requires a pocket_resiudue_cutoff'
AssertionError: distance pocket requires a pocket_resiudue_cutoff

To solve this, I needed to set a --pocket_residue_cutoff 8 in order for the examples to run. Out of curisoity is 8 the minimum cutoff value? As when I try to set it lower it seems to fail.

Many Thanks,
Mike

Hi Hannes,

There is an error when running the test code (the recently released version) of FlowSite as follows:

RuntimeError: Error(s) in loading state_dict for FlowSiteModule:
        Missing key(s) in state_dict: "model.inv_embedder.virtual_atoms"

I think this may be caused by the modification of model codes in the recently released version.
And the first version of your code works well.

I can't find data/inference_csv_example.csv. Am I missing something?

Hi Stark,

Thank you for the amazing work! I have been using your work for my job of redesigning an enzyme active site. I watched your youtube video explaining FlowSite and that was helpful too but I am having trouble interpreting the result since there is no output structure to understand the conformation of the ligand and the residues around it. In the youtube video and the paper you had some figures that illustrates the output therefore I am asking how did you come up with that?