(Personal/Development version)

• Description
• Building and setup instructions
  • Dependencies
  • Setup and compilation
  • Python setup
  • Matlab setup
• Installing in a conda environment (Python)
• Tutorial

SUSAN is a low-level/mid-level framework for fast Subtomogram Averaging (StA) for CryoEM. It uses susbtacks instead of subtomograms that are cropped on-the-fly from the aligned stacks to reduce the computational complexity and to increase the overall performace of the StA pipeline.

SUSAN was designed to be modular, flexible and fast. It is conformed by two layers:

• Low-level layer: Set of executables that perform the demanding computations. They were written in C++ using a minimal set of dependencies: Eigen, as a header-only mathemetical engine, CUDA for GPU acceleration, and PThreads for lightweight multi-threading. Optionally, they can be built with MPI support to run in multi-node environments.
• Mid-level layer: Set of wrappers to the previous layer that simplify its use and provides a set of non time-critical operations. It is used to create the workflows or pipelines as scripts with wrappers for Matlab and for Python. The Matlab one was designed to complement DYNAMO, while the Python one is provided to enable integration to other pipelines based on this language.

I started the development of SUSAN at the Independent Research Group (Sofja Kovaleskaja) of Dr. Misha Kudryashev at the Department of Structural Biology Max Planck Institute of Biophysics (MPIBP) in Frankfurt am Main, Germany. Currently, I am an ARISE fellow at the Kreshuk group at the European Molecular Biology Laboratory (EMBL) in Heidelberg, Germany. Dr. Misha Kudryashev has a new group at the Max Delbrück Center of Molecular Medicine (MDCMM) in Berlin, Germany.

SUSAN is an Open Source project (AGPLv3.0)

• CUDA.
• gcc.
• cmake.
• git.

• OpenMPI, or equivalent, for multi-node support
• Matlab

We assume that SUSAN will be installed in the LOCAL_SUSAN_PATH folder (LOCAL_SUSAN_PATH can be /home/user/Software/, for example)

1. Install the desired dependencies.
2. Clone SUSAN to LOCAL_SUSAN_PATH:

   cd LOCAL_SUSAN_PATH
   git clone https://github.com/rkms86/SUSAN
   

3. Compile SUSAN:

   mkdir bin
   cd bin
   cmake ../
   make -j
   

   Note: The cmake procedure detects the availabilty of OpenMPI and Matlab and compiles their functionalities accordingly.

Besides the standard libraries, the SUSAN module for Python has only two dependencies: NumPy and Numba. Install them if needed:

• Using conda (or equivalent):

  conda install numpy numba
  

• Using pip

  pip install numpy numba
  

LOCAL_SUSAN_PATH must be added to path first and then the susan module can be imported. On the Python command line, or on a Python script:

import sys
sys.path.insert(1,'LOCAL_SUSAN_PATH')
import susan

After executing the optional steps for the Python module, it can be installed using pip and the provided setup.py file. In the LOCAL_SUSAN_PATH execute:

pip install .

After this step, the module susan should be available to be imported.

LOCAL_SUSAN_PATH must be added to path. On the Matlab command line, or on a Matlab script:

addpath LOCAL_SUSAN_PATH

SUSAN can be built and installed inside a conda environment:

• (Optional) Create a new conda environment and activate it:

  conda create -n susan_env
  conda activate susan_env
  

1. Install the basic packages needed for building and using SUSAN

   conda install -c conda-forge git cmake make cudatoolkit-dev=11 gxx=10 numpy numba
   

• (Optional) Install openmpi:

  conda install -c conda-forge openmpi
  

• (Optional) Install packages to run the examples and tutorials:

  conda install -c conda-forge jupyter scipy matplotlib scikit-image
  

2. Go to the directory LOCAL_SUSAN_PATH, where SUSAN will be compiled. For example, LOCAL_SUSAN_PATH can be ~/Software/:

   cd LOCAL_SUSAN_PATH
   

3. Clone SUSAN, compile it and install it:

   git clone https://github.com/rkms86/SUSAN
   mkdir SUSAN/bin
   cd SUSAN/bin
   cmake ../
   make -j
   pip install ../
   

After these step the module susan should be available on the current environment.

A tutorial is available for Python and Matlab for the mixedCTEM dataset from the EMPIAR-10064. It is assumed that the wget and gunzip commands and the IMOD framework are installed in the system.

1. Download the dataset (uses wget):

   cd LOCAL_SUSAN_PATH/tutorials/empiar_10064/data
   ./download_data.sh
   

2. Create the aligned stacks (uses IMOD):

   ./create_binned_aligned_stacks.sh
   

3. Uncompress the initial reference (uses gunzip):

   cd LOCAL_SUSAN_PATH/tutorials/empiar_10064
   gunzip emd_3420_b4.mrc.gz
   

Depending on the system setup:

• For Matlab use workflow.m.
• For Python use workflow.ipynb (as a Jupyter Notebook).