By Chi Zhang, Hongjia Li, Xiaohua Wan, Xuemei Chen, Jieqing Feng, Fa Zhang. This repository is an official implementation of the paper: Transpicker: a transformer-based particle picking framework for cryoEM images.
A 2D particle picker for cryo-EM micrographs.
- Linux, CUDA>=9.2
- Python>=3.7
In order to run the TransPicker we need to install all required packages.
This can be done by creating a virtual environment with
python3 -m venv envand activating it withsource ./env/bin/activate. Once the virtual Python environment is activated, the required packages can be installed with pip usingpip install -r requirements.txt.
You can also run the conda env create -f transpicker_environment.yml to restore the environment.
To get a complete description of usage execute
transpicker -h
You can download datasets from EMPIAR or using your own dataset and organize them as following:
code_root/
└── data/
└── empiarxxxxx/
├── micrographs/
├── 0001.mrc
├── 0002.mrc
└── xxxx.mrc
└── annots/
├── 0001.star
├── 0002.star
└── xxxx.star
You can name your own dataset in other ways, but the micrographs and annots sub directory should be metained (or change the source code).
Then you can run python make_coco_dataset.py in the 'src/transpicker' path to get coco-style datasets as following:
code_root/
└── data/
└── empiarxxxxx/
├── micrographs/
├── 0001.mrc
├── 0002.mrc
└── xxxx.mrc
└── annots/
├── 0001.star(or .box or .txt)
├── 0002.star
└── xxxx.star
└── annotations/
├── instances_train.json
└── instances_val.json
Before training a particle-picking model, you'd better preprocess your datasets using the preprocess.py script.
The preprocess step can be run by:
python preprocess.py
Available optionals:
usage: Preprocessing script [-h] [--split SPLIT]
[--is_equal_hist IS_EQUAL_HIST]
[--denoise_model {n2n,lowpass,gaussian,nlm,bi_filter}]
[--root_dir ROOT_DIR] [--split_num SPLIT_NUM]
[--split_gap SPLIT_GAP]
[--iou_threshold IOU_THRESHOLD] [--ext EXT]
optional arguments:
-h, --help show this help message and exit
--split SPLIT If need split the micrographs and the responding
annotations. No more than 200 particles are
recommended for each micrograph patch.
--is_equal_hist IS_EQUAL_HIST
If need do histogram equalization. Default is True.
--denoise_model {n2n,lowpass,gaussian,nlm,bi_filter}
Choose a denoise model.
--root_dir ROOT_DIR Path to dataset.
--split_num SPLIT_NUM
The number of patches you want to split in each row
and column.
--split_gap SPLIT_GAP
The overlap that needs to be left for segmentation.
The recommended size of the interval is slightly
larger than the particle diameter.
--iou_threshold IOU_THRESHOLD
bbox less than this threshold will not be saved.
--ext EXT The extention of micrograph file type.
The training step can be run by:
cd bin
sh train.sh
You can use -h to view all parameters.
The particle prediction can be run by
python predict.py
If you want to use GUI to display the micrographs and choose better thresholds, you can use:
python boxmanager.py
to activate the GUI panel.
If you find TransPicker useful in your research, please consider citing:
@inproceedings{zhang2021transpicker,
title={TransPicker: a transformer-based framework for particle picking in cryoEM micrographs},
author={Zhang, Chi and Li, Hongjia and Wan, Xiaohua and Chen, Xuemei and Yang, Zhenghe and Feng, Jieqing and Zhang, Fa},
booktitle={2021 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)},
pages={1179--1184},
year={2021},
organization={IEEE}
}
If you have any questions or require any further information, welcome to contact me.
Email: [email protected]