Deep learning for cryo-EM micrograph cleaning
MicrographCleaner (micrograph_cleaner_em) is a python package designed to segment cryo-EM micrographs into:
- carbon/high-contrast or contaminated regions
- good regions
so that incorrectly picked coordinates can be easily ruled out
To get a complete description of usage execute
cleanMics -c path/to/inputCoords/ -o path/to/outputCoords/ -b 180 -s 1.0 -i /path/to/micrographs/ --predictedMaskDir path/to/store/masks --deepThr 0.5
Install anaconda Python 3x version from https://www.anaconda.com/distribution/
Create an environment for MicrographCleaner
conda create -n env_micrograph_cleaner_em python=3.6
Activate environment (each time you want to use micrograph_cleaner you will need to activate it)
conda activate env_micrograph_cleaner_em
Install micrograph_cleaner_em from repository
conda install -c rsanchez1369 micrograph-cleaner-em
Download deep learning model
- install CUDA and cudnn in such a way that tensorflow (https://www.tensorflow.org/) can be executed.
micrograph_cleaner is compatible with CUDA-8,CUDA-9 and CUDA-10.
Tensorflow version will be automatically selected according your CUDA version and installed later.
CUDA is available at https://developer.nvidia.com/cuda-toolkit and cudnn is available at
Easy cudnn instalation can be performed automatically at step 2 using python module cudnnenv
1.1) (optional) create virtual environment
pip install virtualenv virtualenv --system-site-packages -p python3 ./env_micrograph_cleaner_em source ./env_micrograph_cleaner_em/bin/activate
- Install micrograph_cleaner_em
git clone https://github.com/rsanchezgarc/micrograph_cleaner_em.git cd micrograph_cleaner_em python setup.py install
pip install micrograph_cleaner_em
2.1) If cudnn not installed yet, install install cudnnenv
pip install cudnnenv
cudnnenv install [VERSION], where recommended versions are "v6-cuda8" for CUDA-8, "v7.0.1-cuda9" for CUDA-9 and
"v7.4.1-cuda10" for CUDA-10.
Download deep learning model
Install scipion version 2.0+ from http://scipion.i2pc.es/
Install xmipp either from plugin manager or from command line
scipion installp -p scipion-em-xmipp
Install deepLearningToolkit either from plugin manager or from command line
scipion installb deepLearningToolkit
MicrographCleaner employs an U-net-based deep learning model to segmentate micrographs into good regions and bad regions. Thus, it is mainly used as a post-processing step after particle picking in which coordinates selected in high contrast artefacts, such as carbon, will be ruled out. Additionally, it can be employed to generate binary masks so that particle pickers can be prevented from considering problematic regions. Thus, micrograph_cleaner employs as a mandatory argument a(some) micrograph(s) fileneame(s) and the particle size in pixels (with respect input mics). Additionally it can recive as input:
- A directory where picked coordinates are located and another directory where scored/cleaned coordiantes will be saved. Coordinates will be saved in pos format or plain text (columns whith header colnames x and y) are located.
There must be one different coordinates file for each micrograph named as the micrograph and the output coordiantes will preserve the naming.
E.g. -c path/to/inputCoordsDirectory/ -o /path/to/outputCoordsDirectory/
Allowed formats are xmipp pos and raw text tab separated with at least two columns named as xcoor, ycoor in the header. Raw text file example:
micFname1.tab: ########################################### xcoor ycoor otherInfo1 otherInfo2 12 143 -1 0.1 431 4341 0 0.2 323 321 1 0.213 ###########################################
A directory where predicted masks will be saved (mrc format). E.g. --predictedMaskDir path/where/predictedMasksWillBeSaved/
A downsampling factor (can be less than 1 if actually upsampling was performed) in case the coordinates where picked from micrographs at different scale.
E.g. -s 2 will downsample coordinates by a factor 2 and then it will apply the predicted mask that is as big as the input micrographs. This case corresponds to an example in which we use for particle picking raw micrographs but we are using MicrographCleaner with downsampled mics
Any combination of previous options.
Trained MicrographCleaner model is available at http://campins.cnb.csic.es/micrograph_cleaner/ and can be automatically download executing
Beware that if you installed micrograph_cleaner using pip/source, then CUDA and cudnn libraries should be available prior execution, so if CUDA is not found, export its path prior execution
and then execute
#Donwload deep learning model cleanMics --download #Compute masks from imput micrographs and store them cleanMics -b $BOX_SIXE -i /path/to/micrographs/ --predictedMaskDir path/to/store/masks #Rule out input bad coordinates (threshold<0.5) and store them into path/to/outputCoords cleanMics -c path/to/inputCoords/ -o path/to/outputCoords/ -b $BOX_SIXE -s $DOWN_FACTOR -i /path/to/micrographs/ --deepThr 0.5 #Compute goodness scores from input coordinates and store them into path/to/outputCoords cleanMics -c path/to/inputCoords/ -o path/to/outputCoords/ -b $BOX_SIXE -s $DOWN_FACTOR -i /path/to/micrographs/ --deepThr 0.5
The fundamental class employed within MicrographCleaner is MaskPredictor, a class designed to predict a contamination/carbon mask given a micrograph.
Usage: predicts masks of shape HxW given one numpy array of shape that represents a micrograph. Mask values range from 0. to 1., being 0. associated to clean regions and 1. to contamination.
micrograph_cleaner_em.MaskPredictor(deepLearningModelFname, boxSize , gpus=, strideFactor=2) :param deepLearningModelFname (str): a path where the deep learning model will be loaded :param boxSize (int): estimated particle boxSize in pixels :param gpus (list of gpu ids (ints) or None): If None, CPU only mode will be employed. :param strideFactor (int): Overlapping between windows. Micrographs are divided into patches and each processed individually. The overlapping factor indicates how many times a given row/column is processed by the network. The bigger the better the predictions, but higher computational cost.
predictMask(self, inputMic): Obtains a contamination mask for a given inputMic :param inputMic (np.array shape HxW): the micrograph to clean :return: mask (np.array shape HxW): a mask that ranges from 0. to 1. -> 0. meaning clean area and 1. contaminated area.
getDownFactor(self): MaskPredictor preprocess micrographs before Nnet computation. First step is donwsampling using a donwsampling factor that depends on particle boxSize. This function computes the donwsampling factor :return (float): the donwsampling factor that MaskPredictor uses internally when preprocessing the micrographs close(self): Used to release memory
The following lines show how to compute the mask for a given micrograph
from micrograph_cleaner_em import MaskPredictor deepLearningModelFname="/path/to/deepLearningModel.keras" boxSize=128 mic= loadMic("do whatever you need") # mic is a np.array with MaskPredictor(deepLearningModelFname, boxSize , gpus=) as mp: mask= mp.predict(mic)
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size micrograph_cleaner_em-0.22.tar.gz (17.6 kB)||File type Source||Python version None||Upload date||Hashes View|
Hashes for micrograph_cleaner_em-0.22.tar.gz