DeepEMhancer

Simply speaking, DeepEMhancer performs a non-linear post-processing of cryo-EM maps that produces two main effects:
1. Local sharpening-like post-processing.
2. Automatic masking/denoising of cryo-EM maps.

git clone https://github.com/yyuu/pyenv.git /apps/pyenv
export PYENV_ROOT=/apps/pyenv
export PATH=$PYENV_ROOT/bin:$PATH
pyenv install anaconda3-2023.07-2
pyenv global anaconda3-2023.07-2
export PATH=$PYENV_ROOT/versions/anaconda3-2023.07-2/bin/:$PATH

export PYENV_ROOT=/apps/pyenv
export PATH=$PYENV_ROOT/bin:$PATH
export PATH=$PYENV_ROOT/versions/anaconda3-2023.07-2/bin/:$PATH

[root@rockylinux ~]# cd /apps/
[root@rockylinux apps]# git clone https://github.com/rsanchezgarc/deepEMhancer
[root@rockylinux apps]# cd deepEMhancer
[root@rockylinux deepEMhancer]# ls
alternative_installation  condaDeepEMHancer  conda_ymls  deepEMhancer  deepEMhancer_env.yml  deepEMhancer_tutorial_06_23.pdf  LICENSE  README.md  setup.py
 
[root@rockylinux deepEMhancer]# conda env create -f deepEMhancer_env.yml  -n deepEMhancer_env
Collecting package metadata (repodata.json): done
Solving environment: \
 :
 :
done
#
# To activate this environment, use
#
#     $ conda activate deepEMhancer_env
#
# To deactivate an active environment, use
#
#     $ conda deactivate
 
[root@rockylinux deepEMhancer]#

[root@rockylinux deepEMhancer]# source activate deepEMhancer_env
(deepEMhancer_env) [root@rockylinux deepEMhancer]# conda list
 :
cudatoolkit               11.8.0              h4ba93d1_12    conda-forge
cudnn                     8.8.0.121            h0800d71_1    conda-forge
 :
tensorboard               2.12.3                   pypi_0    pypi
tensorboard-data-server   0.7.1                    pypi_0    pypi
tensorflow                2.12.0                   pypi_0    pypi
tensorflow-estimator      2.12.0                   pypi_0    pypi
tensorflow-io-gcs-filesystem 0.32.0                   pypi_0    pypi
 :
(deepEMhancer_env) [root@rockylinux deepEMhancer]# pwd
/apps/deepEMhancer
 
(deepEMhancer_env) [root@rockylinux deepEMhancer]# which python
/apps/pyenv/versions/anaconda3-2023.07-2/envs/deepEMhancer_env/bin/python
 
(deepEMhancer_env) [root@rockylinux deepEMhancer]# 
(deepEMhancer_env) [root@rockylinux deepEMhancer]# python -m pip install . --no-deps
Looking in indexes: https://pypi.org/simple, https://pypi.ngc.nvidia.com
Processing /apps/deepEMhancer
  Preparing metadata (setup.py) ... done
Building wheels for collected packages: deepEMhancer
  Building wheel for deepEMhancer (setup.py) ... done
  Created wheel for deepEMhancer: filename=deepEMhancer-0.16-py3-none-any.whl size=35142 sha256=c586c01e29a4cfbc3c461f4b30c06c78afc66dd653a98b185030f17bedcf8201
  Stored in directory: /tmp/pip-ephem-wheel-cache-wsj3carg/wheels/cf/c6/c8/1857699bf687d7bd754dd9dad454c30896add113827c5ed02a
Successfully built deepEMhancer
Installing collected packages: deepEMhancer
Successfully installed deepEMhancer-0.16
WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv
 
(deepEMhancer_env) [root@rockylinux deepEMhancer]# conda deactivate
[root@rockylinux deepEMhancer]#

[root@rockylinux ~]# vi /apps/modulefiles/deepEMhancer
#%Module1.0
set          root       /apps/pyenv/versions/anaconda3-2023.03/envs/deepEMhancer_env/
prepend-path PATH       $root/bin
prepend-path LD_LIBRARY_PATH $root/lib:$root/lib/python3.9/site-packages/nvidia/cudnn/lib
 
[root@rockylinux ~]#

[saber@rockylinux ~]$ module use /apps/modulefiles/
[saber@rockylinux ~]$ module load deepEMhancer
 
[saber@rockylinux ~]$ deepemhancer --download
DOWNLAODING MODELs from https://zenodo.org/record/7432763/files/deepEMhancerModels_tf2.zip to /home/saber/.local/share/deepEMhancerModels
It may take a while...
100%|???????????????????????????????????????????????????????????????????????????????????????????????????????????| 180462/180464 [01:50<00:00, 1630.69it/s]
Total size downloaded: 739172352.
Unzipping...
DONE!!
 
[saber@rockylinux ~]$ ls -lh .local/share/deepEMhancerModels/production_checkpoints/
total 884M
-rw-rw-r--. 1 saber saber 298M May 22 02:22 deepEMhancer_highRes.hd5
-rw-rw-r--. 1 saber saber 196M May 22 02:22 deepEMhancer_masked.hd5
-rw-rw-r--. 1 saber saber 196M May 22 02:22 deepEMhancer_tightTarget.hd5
-rw-rw-r--. 1 saber saber 196M May 22 02:22 deepEMhancer_wideTarget.hd5
-rw-rw-r--. 1 saber saber   74 May 22 02:22 versions.txt
[saber@rockylinux ~]$
[saber@rockylinux ~]$ cat .local/share/deepEMhancerModels/production_checkpoints/versions.txt
highRes: g2_3
tightTarget: v3_L_c_2
wideModel: v3t_28_c+
masked: vt2_m_28
 
[saber@rockylinux ~]$

[saber@rockylinux ~]$ deepemhancer -h
usage: deepemhancer -i INPUTMAP -o OUTPUTMAP [-p {wideTarget,tightTarget,highRes}] [-i2 HALFMAP2] [-s SAMPLINGRATE] [--noiseStats NOISE_MEAN NOISE_STD]
                    [-m BINARYMASK] [--deepLearningModelPath PATH_TO_MODELS_DIR] [--cleaningStrengh CLEANINGSTRENGH] [-g GPUIDS] [-b BATCH_SIZE]
                    [--version] [-h] [--download [DOWNLOAD_DEST]]
 
DeepEMHancer. Deep post-processing of cryo-EM maps. https://github.com/rsanchezgarc/deepEMhancer
 
options:
  -h, --help            show this help message and exit
  --download [DOWNLOAD_DEST]
                        download default DeepEMhancer models. They will be saved at /home/saber/.local/share/deepEMhancerModels/production_checkpoints
                        if no path provided
 
Main options:
  -i INPUTMAP, --inputMap INPUTMAP
                        Input map to process or half map number 1. This map should be unmasked and not sharpened (Do not use post-processed maps, only
                        maps directly obtained from refinement). If half map 1 used, do not forget to also provide the half map 2 using -i2
  -o OUTPUTMAP, --outputMap OUTPUTMAP
                        Output fname where post-processed map will be saved
  -p {wideTarget,tightTarget,highRes}, --processingType {wideTarget,tightTarget,highRes}
                        Select the deep learning model you want to use. WideTarget will produce less sharp results than tightTarget. HighRes is only
                        recommended for overal FSC resolution < 4 A This option is igonred if normalization mode 2 is selected
  -i2 HALFMAP2, --halfMap2 HALFMAP2
                        (Optional) Input half map 2 to process
  -s SAMPLINGRATE, --samplingRate SAMPLINGRATE
                        (Optional) Sampling rate (A/voxel) of the input map. If not provided, the sampling rate will be read from mrc file header
 
Normalization options (auto normalization is applied if no option selected):
  --noiseStats NOISE_MEAN NOISE_STD
                        (Optional) Normalization mode 1: The statisitcs of the noise to normalize (mean and standard deviation) the input. Preferred
                        over binaryMask but ignored if binaryMask provided. If not --noiseStats nor --binaryMask provided, nomralization params will be
                        automatically estimated, although, in some rare cases, estimation may fail or be less accurate
  -m BINARYMASK, --binaryMask BINARYMASK
                        (Optional) Normalization mode 2: A binaryMask (1 protein, 0 no protein) used to normalize the input. If no normalization mode
                        provided, automatic normalization will be carried out. Supresses --precomputedModel option
 
Alternative options:
  --deepLearningModelPath PATH_TO_MODELS_DIR
                        (Optional) Directory where a non default deep learning model is located (model is selected using --precomputedModel) or a path
                        to hd5 file containing the model
  --cleaningStrengh CLEANINGSTRENGH
                        (Optional) Post-processing step to remove small connected components (hide dust). Max relative size of connected components to
                        remove 0<s<1 or -1 to deactivate. Default: -1
 
Computing devices options:
  -g GPUIDS, --gpuIds GPUIDS
                        The gpu(s) where the program will be executed. If more that 1, comma seppared. E.g -g 1,2,3. Set to -1 to use only cpu (very
                        slow). Default: 0
  -b BATCH_SIZE, --batch_size BATCH_SIZE
                        Number of cubes to process simultaneously. Lower it if CUDA Out Of Memory error happens and increase it if low GPU performance
                        observed. Warning, for some inputs it may crash if --gpus > . Use only 1 gpus in that case. 1Default: 8
  --version             show program's version number and exit
 
examples:
 
  + Download deep learning models
deepemhancer --download
 
  + Post-process input map path/to/inputVol.mrc and save it at path/to/outputVol.mrc using default  deep model tightTarget
deepemhancer  -i path/to/inputVol.mrc -o  path/to/outputVol.mrc
 
  + Post-process input map path/to/inputVol.mrc and save it at path/to/outputVol.mrc using high resolution deep model
deepemhancer -p highRes -i path/to/inputVol.mrc -o  path/to/outputVol.mrc
 
  + Post-process input map path/to/inputVol.mrc and save it at path/to/outputVol.mrc using a deep learning model located in path/to/deep/learningModel
deepemhancer -c path/to/deep/learningModel -i path/to/inputVol.mrc -o  path/to/outputVol.mrc
 
  + Post-process input map path/to/inputVol.mrc and save it at path/to/outputVol.mrc using high resolution  deep model and providing normalization information (mean
    and standard deviation of the noise)
deepemhancer -p highRes -i path/to/inputVol.mrc -o  path/to/outputVol.mrc --noiseStats 0.12 0.03
[saber@rockylinux ~]$

[saber@rockylinux test]$ export LD_LIBRARY_PATH=/apps/pyenv/versions/anaconda3-2023.07-2/envs/deepEMhancer_env/lib/python3.9/site-packages/nvidia/cudnn/lib    <--- これが有効でないとGPUを使わないみたい
 
[saber@rockylinux test]$ deepemhancer -i ../relion40_tutorial_precalculated_results/Refine3D/job029/run_it000_half1_class001.mrc -i2 ../relion40_tutorial_precalculated_results/Refine3D/job029/run_it000_half2_class001.mrc \
-p tightTarget -o model.mrc --deepLearningModelPath  ~/.local/share/deepEMhancerModels/production_checkpoints/   -g 0,1,2,3
 
updating environment to select gpu: [0, 1, 2, 3]
loading model /apps/deepEMhancer/deepEMhancerModels/production_checkpoints/deepEMhancer_tightTarget.hd5 ... DONE!
Automatic radial noise detected beyond 36 % of volume side
DONE!. Shape at 1.00 A/voxel after padding->  (320, 320, 320)
Neural net inference
100%|?????????????????????????????????????????????????????????????????????????????????????????????????????????????????| 289/289 [4:58:14<00:00, 61.92s/it]
[saber@rockylinux test]$