biomap-research / xtrimomultimer Goto Github PK

When i use bin/inference.sh to test the code, the erro reported that cannot find such npz file, and i cannot find it in github as well
report like this:
-- Process 0 terminated with the following error:
Traceback (most recent call last):
File "/data/user/liwh/.conda/envs/xTrimoMultimer/lib/python3.9/site-packages/torch/multiprocessing/spawn.py", line 69, in wrap
fn(i, *args)
File "/data/user/liwh/xTrimoMultimer/inference.py", line 371, in predict_structure
import_jax_weights(model, param_path, version=model_name)
File "/data/user/liwh/xTrimoMultimer/xtrimomultimer/utils/import_weights.py", line 550, in import_jax_weights_
data = np.load(npz_path)
File "/data/user/liwh/.conda/envs/xTrimoMultimer/lib/python3.9/site-packages/numpy/lib/npyio.py", line 405, in load
fid = stack.enter_context(open(os_fspath(file), "rb"))
FileNotFoundError: [Errno 2] No such file or directory: './data/params/params_model_1.npz'

Is there someone experienced who can tell me what's going on
thanks so much

it failed to process features model. could anyone give a suggestion?

Traceback (most recent call last):
File "inference.py", line 657, in
main(args)
File "inference.py", line 566, in main
is_multimer=global_is_multimer,
File "/data/xTrimoMultimer/xtrimomultimer/data/feature_pipeline.py", line 108, in process_features
mode=mode,
File "/data/xTrimoMultimer/xtrimomultimer/data/feature_pipeline.py", line 77, in np_example_to_features
num_res = int(np_example["seq_length"])
TypeError: only size-1 arrays can be converted to Python scalars

Just cloned the project and was giving a try of inference and I ran into this:

Traceback (most recent call last):
  File "inference.py", line 658, in <module>
    main(args)
  File "inference.py", line 535, in main
    alignment_runner.run(chain_fasta_path, local_alignment_dir)
  File "/xTrimoMultimer/xtrimomultimer/data/data_pipeline.py", line 476, in run
    if self.template_searcher.input_format == "sto":
AttributeError: 'HHSearch' object has no attribute 'input_format'

So I took a look at this HHSearch object at data/tools/hhsearch.py. Turns out that it doesn't have a input_format property. So I tried to seek this property name within the package and found that it is provided in a certain Hmmsearch class. So it seems that these 2 wrappers are so alike that the HHSearch class should've had the same properties and methods but they were somehow forgotten.

Building the docker image on Alma 8, build fails

$ docker buildx build . --file Dockerfile --tag xtrimomultimer:latest
...
#0 398.5 Linking aiohttp-3.8.3-py37h540881e_0
#0 398.6 Linking google-auth-1.35.0-pyh6c4a22f_0
#0 398.6 Linking google-auth-oauthlib-0.4.6-pyhd8ed1ab_0
#0 398.6 Linking tensorboard-2.4.1-pyhd8ed1ab_1
#0 398.8 Linking pytorch-lightning-1.5.10-pyhd8ed1ab_0
#0 399.1 Transaction finished
#0 399.1 
#0 399.1 Installing pip packages: deepspeed==0.5.3, ml-collections==0.1.0, colossalai
#0 400.3 Collecting deepspeed==0.5.3
#0 401.0   Downloading deepspeed-0.5.3.tar.gz (477 kB)
#0 401.2      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 477.5/477.5 kB 4.4 MB/s eta 0:00:00
#0 401.3   Preparing metadata (setup.py): started
#0 403.4   Preparing metadata (setup.py): finished with status 'done'
#0 403.5 Collecting ml-collections==0.1.0
#0 403.5   Downloading ml_collections-0.1.0-py3-none-any.whl (88 kB)
#0 403.5      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 88.7/88.7 kB 6.4 MB/s eta 0:00:00
#0 403.6 ERROR: Could not find a version that satisfies the requirement colossalai (from versions: none)
#0 403.6 ERROR: No matching distribution found for colossalai
#0 404.1 critical libmamba pip failed to install packages
------
ERROR: failed to solve: executor failed running [/bin/sh -c micromamba create -n xtrimomultimer -f /opt/xTrimoMultimer/requirements/environment.yaml -y  && micromamba clean --all]: exit code: 1

https://colossalai.org/docs/get_started/installation/ was last updated 17-Oct-22, looks like they've done away with the vanilla pip install

gammut@host2[~]:pip3 install --user colossalai
ERROR: Could not find a version that satisfies the requirement colossalai (from versions: none)
ERROR: No matching distribution found for colossalai

program stopped at init_dap() when using --gpu_nums 2, after comment list.all_gather_object line, it runs.

ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (5,) + inhomogeneous part.

• inference.sh

python inference.py /data/scratch/alphafold/alphafold//pdb_mmcif/mmcif_files/
--fasta_paths target.fasta
--model_preset multimer
--output_dir ./tmp/
--param_dir /data/scratch/alphafold/alphafold//params/
--gpu_nums 1
--cpus 12
--use_fastfold_optimize
--uniref90_database_path /data/scratch/alphafold/alphafold//uniref90/uniref90.fasta
--mgnify_database_path /data/scratch/alphafold/alphafold//mgnify/mgy_clusters_2018_12.fa
--pdb70_database_path /data/scratch/alphafold/alphafold//pdb70/pdb70
--uniclust30_database_path /data/scratch/alphafold/alphafold//uniclust30/uniclust30_2018_08/uniclust30_2018_08
--pdb_seqres_database_path /home/lcmql/data/pdb_seqres.txt
--bfd_database_path /data/scratch/alphafold/alphafold/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt
--jackhmmer_binary_path which jackhmmer
--hhblits_binary_path which hhblits
--hhsearch_binary_path which hhsearch
--hmmsearch_binary_path which hmmsearch
--kalign_binary_path which kalign
--hmmbuild_binary_path which hmmbuild
--model_name model_1_multimer,model_2_multimer,model_3_multimer,model_4_multimer,model_5_multimer

Can you share the details of the following benchmark? Is that an end-to-end benchmark? With A100 40G? Multimer or single chain? (as OpenFold didn't support multimer yet). Any scripts to reproduce the benchmark results?
How about the results of single GPU of xTrimo-Multimer?

Thanks for the code!

I read an interview of Prof. Le Song, and he said that Xtrimo can predict the reaction of a single cell to outer stimulation. My question is how can I do this with the Xtrimo code?

Thanks in advance!

It surprised me that the multimer version from Baidu is not based on helixfold(https://github.com/PaddlePaddle/PaddleHelix/tree/dev/apps/protein_folding/helixfold), but OpenFold. Especially the technologies are very similar, such as Dynamic Axial Parallelism (DAP) and kernel fusion.

python /inference.py --help

Python 3.7.12
Traceback (most recent call last):
File "/opt/xTrimoMultimer/inference.py", line 34, in
import xtrimomultimer.np.relax.relax as relax
File "/opt/xTrimoMultimer/xtrimomultimer/np/relax/init.py", line 11, in
_modules = [(m, importlib.import_module("." + m, name)) for m in all]
File "/opt/xTrimoMultimer/xtrimomultimer/np/relax/init.py", line 11, in
_modules = [(m, importlib.import_module("." + m, name)) for m in all]
File "/opt/conda/envs/xtrimomultimer/lib/python3.7/importlib/init.py", line 127, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "/opt/xTrimoMultimer/xtrimomultimer/np/relax/relax.py", line 19, in
from xtrimomultimer.np.relax import amber_minimize, utils
File "/opt/xTrimoMultimer/xtrimomultimer/np/relax/amber_minimize.py", line 23, in
import xtrimomultimer.utils.loss as loss
File "/opt/xTrimoMultimer/xtrimomultimer/utils/init.py", line 2, in
from xtrimomultimer.utils.rich_utils import enforce_tags, print_config_tree
ModuleNotFoundError: No module named 'xtrimomultimer.utils.rich_utils'

https://github.com/biomap-research/xTrimoMultimer/tree/main/xtrimomultimer/utils
cann't find rich_utils.py and rich_utils.py