Transpicker

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.

Introduction

A 2D particle picker for cryoEM micrographs.

Abstract

License

Citing TransPicker

If you find TransPicker useful in your research, please consider citing:

Installation

Requirements

• 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 env and activating it with source ./env/bin/activate. Once the virtual Python environment is activated, the required packages can be installed with pip using pip install -r requirements.txt.

You can also run the conda env create -f transpicker_environment.yml to restore the environment.

Usage

To get a complete description of usage execute

transpicker -h

Dataset preparation

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).

Make coco-style dataset for training and testing

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

Preprocessing

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.

Training

The training step can be run by:

cd bin \ sh train.sh

You can use -h to view all parameters.

Prediction

The particle prediction can be run by

python predict.py

GUI usage

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.

Contact

If you have any questions or require any further information, welcome to contact me.

Email: chizhang_cs@zju.edu.cn