pcnaDeep: a deep-learning based single-cell cycle profiler with PCNA signal

Welcome! pcnaDeep integrates cutting-edge detection techniques with tracking and cell cycle resolving models. Using the Mask R-CNN model under FAIR's Detectron2 framework, pcnaDeep is able to detect and resolve very dense cell tracks with PCNA fluorescence.

Installation

1. PyTorch (torch >= 1.7.1) installation and CUDA GPU support are essential. Visit PyTorch homepage for specific installation schedule.
2. Install modified Detectron2 v0.4 in this directory (original package homepage)

      cd detectron2-04_mod
      pip install .
   

   • In pcnaDeep, the detectron2 v0.4 dependency has been modified in two ways:
     1. To generate confidence score output of the instance classification, the method detectron2.modeling.roi_heads.fast_rcnn.fast_rcnn_inference_single_image has been modified.
     2. A customized dataset mapper function has been implemented as detectron2.data.dataset_mapper.read_PCNA_training.
   • To build Detectron2 on Windows may require the following change of torch package, if your torch version is old. Reference (Chinese).

      In torch\include\torch\csrc\jit\argument_spec.h,
      static constexpr size_t DEPTH_LIMIT = 128;
         change to -->
      static const size_t DEPTH_LIMIT = 128;
   

3. Install pcnaDeep from source in this directory

   cd bin
   python setup.py install
   

4. (optional, for training data annotation only) Download VGG Image Annotator 2 software.
5. (optional, for visualisation only) Install Fiji (ImageJ) with TrackMate CSV Importer plugin.

Download pre-trained Mask R-CNN weights

The Mask R-CNN is trained on 60X microscopic images sized 1200X1200 square pixels. Download here.

You must download pre-trained weights and save it under ~/models/ for running tutorials.

Getting started

See a quick tutorial to get familiar with pcnaDeep.

You may also go through other tutorials for advanced usages.

API Documentation

API documentation is available here.

Reference

Please cite our paper if you found this package useful.

pcnaDeep: A Fast and Robust Single-Cell Tracking Method Using Deep-Learning Mediated Cell Cycle Profiling
Yifan Gui, Shuangshuang Xie, Yanan Wang, Ping Wang, Renzhi Yao, Xukai Gao, Yutian Dong, Gaoang Wang, Kuan Yoow Chan
bioRxiv 2021.09.19.460933; doi: https://doi.org/10.1101/2021.09.19.460933

Licence

pcnaDeep is released under the Apache 2.0 license.