DeePiCt

Source code for the paper:

Convolutional networks for supervised mining of molecular patterns within cellular context. In preparation.

de Teresa, I.*, Goetz S.K.*, Mattausch, A., Stojanovska, F., Zimmerli C., Toro-Nahuelpan M., Cheng, D.W.C., Tollervey, F. , Pape, C., Beck, M., Kreshuk, A., Mahamid, J. and Zaugg, J.

Table of Contents

1. Introduction
2. Installation
3. How to run
4. Colab Notebooks
5. Trained Models
6. Useful_Scripts

1. Introduction

With improved instrumentation and sample preparation protocols, a large number of high-quality cryo-ET images are rapidly being generated in laboratories, opening the possibility to conduct high-throughput studies in cryo-ET. However, due to the crowded nature of the cellular environment together with image acquisition limitations, data mining and image analysis of cryo-ET tomograms remains one of the most important bottlenecks in the field. We present DeePiCt (Deep Picker in Context), a deep-learning based pipeline to achieve structure segmentation and particle localization in cryo-electron tomography. DeePiCt combines two dedicated convolutional networks: a 2D CNN for segmentation of cellular compartments (e.g. organelles or cytosol), and a 3D CNN for particle localization and structure segmentation.

Figure 1 | DeePiCt's Workflow for Segmentation of cellular structures. a. Both the 2D CNN and the 3D CNN for DeePiCt workflow are variations of the U-Net architecture (Ronnenberg et al., 2015). b. An example of DeePict's workflow for the segmentation of membranes and the localization of fatty-acid synthases (FAS) and cytosolic ribosomes in a S. pombe cryo-tomogram.

2. Installation

Both 2D and 3D CNN pipelines require a conda installation, and are run via the Snakemake workflow management system.

Requirements and conda environment

Package Installation (miniconda, Pytorch and Keras).

Miniconda

Download the latest miniconda3 release, according to your OS and processor (modify the Miniconda3-latest-Linux-x86_64.sh file according to the instructions available here):

cd foldertodownload
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

during installation, you'll be asked where to install miniconda. Select a folder with large capacity:

/path/to/folder/with/large/capacity/miniconda3

Virtual environment

Create a basic conda environment with Snakemake and pandas

The necessary virtual environment for the installation needs only snakemake and pandas

conda install -n base -c conda-forge mamba
conda activate base
mamba create -c conda-forge -c bioconda -n snakemake snakemake==5.13.0 python=3.8
conda activate snakemake
conda install pandas

Install Pytorch:

conda install -c pytorch pytorch-gpu torchvision

Install Keras:

conda install -c conda-forge keras-gpu=2.3.1

Clone this repository

cd /folder/where/the/repository/will/be/cloned
git clone https://github.com/ZauggGroup/DeePiCt.git

3. How to run

Go to the folder where you plan to run the experiments. Create a configuration file -with the structure given in the examples (see each of the 2d_cnn/config.yaml or 3d_cnn/config.yaml files)-. Run the pipeline by:

• 2D CNN pipeline: bash /path/to/2d_cnn/deploy_cluster.sh /path/to/config.yaml

• 3D CNN pipeline: bash /path/to/3d_cnn/deploy_cluster.sh /path/to/config.yaml

(notice that, to run locally the deploy_cluster.sh script can be exchanged by deploy_local.sh).

Configuration

We refer to the 2d_cnn/README.md and 3d_cnn/README.md files for corresponding specifications.

4. Colab Notebooks

We provide two notebooks to try out the prediction of 2D and 3D CNN trained models on one tomogram. The the spectrum matching filter is not included in the notebooks. This step should be done beforehand, following the instructions here.

To predict cytosol or organelles, you can use the 2D trained models:

DeePiCt_predict2d.ipynb .

To predict ribosome, membrane, microtubules or FAS, you can use the 3D trained models:

DeePiCt_predict3d.ipynb .

5. Trained Models

Trained models are available here. All models were trained with cryo-ET data (4-times binned, unbinned pixel size 3.37 A) pre-processed using the spectrum matching filter with spectrum_TS_001.tsv.

6. Additional Scripts

A number of useful scripts can be found in the folder additional_scripts/. python additional_scripts/<script_name> --help to learn how to use it.

Below the list.

• motl2sph_mask.py

Script that converts coordinate lists into spherical masks, to produce training data for the 3D CNN. Example:

python DeePiCt/additional_scripts/motl2sph_mask.py -r 8 -motl test_motl.csv -o \
Downloads/test_mask.mrc -shape 900 900 500 -value 1

• elliptical_distance_constraint.py

Script to merge several lists of coordinates into a single one, avoiding duplicates and imposing elliptical distance constrains to respect (possibly different) minimal distance between points along axis x, y, z. The elliptic coefficients a b and c represent the corresponding minimum distance in voxels. Example:

python DeePiCt/additional_scripts/elliptical_distance_constraint.py --abc 9 9 15 -f test_motl1.csv test_motl2.csv \
-o merged_list.csv