MemXTerminator

A software for membrane analysis and subtraction in cryo-EM.

Overview

This software utilizes 2D averages and their corresponding alignment information, employing methods such as Radon transform, cross-correlation, L1 norm, Bezier curves, Monte Carlo simulations, and Genetic Algorithm. It analyzes and subtracts membranes of any shape in cryo-EM, ultimately producing particle stacks and micrographs with membrane signals removed, which are suitable for subsequent membrane protein analysis.

Features

• Capable of analyzing biological membranes of any shape, including simple lines and arcs, as well as more complex shapes like S or W curves;
• Accurately locates and subtracts biological membrane signals;
• Utilizes GPU and CUDA acceleration to enhance computational speed;
• Features a user-friendly GUI for ease of use.

Requirements

• This software requires a GPU and CUDA acceleration. So, the installation of CUDA drivers and libraries is necessary.
• pyem is also needed to convert cryoSPARC’s .cs files to Relion’s .star format for processing.

Wiki

This software has a very accessible wiki for better tutorial organization.

Installation

For specific installation methods, please refer to the wiki installation section.

Usage

This software has a user-friendly GUI. To use this software, simply type:

MemXTerminator gui &

For detailed usage tutorials, please refer to the wiki usage section.

Fork maintenance notes

This fork keeps the upstream workflow, with a few operational fixes for HPC use:

• The GUI configures Linux X11 sessions before importing Qt: QT_X11_NO_MITSHM=1 is set to avoid fragile shared-memory behavior over SSH forwarding. Matplotlib backend selection is left to the running Qt GUI framework.
• RadonFit membrane analysis accepts --procs/--cpu for template-level multiprocessing. Use --procs 96 on a 96-core node to process up to 96 templates concurrently; 0 auto-detects CPU count and caps at the template count.
• RadonFit particle membrane subtraction treats --procs as GPU worker processes; the default 0 auto-detects visible CUDA devices. Each worker logs its assigned CUDA device at startup.
• --batch_size is now a progress/reporting window; real parallelism is controlled by --procs, avoiding minibatch barriers that can leave GPUs idle.
• RadonFit particle and micrograph membrane subtraction accept --output_dirname; use the same value for both steps so MMS finds the matching PMS stacks and .mxt sidecars.
• On HPC systems where CUDA modules prepend their own Python, install/run from the conda env explicitly, e.g. $CONDA_PREFIX/bin/python -m pip install -e .; GUI-launched jobs prefer $CONDA_PREFIX/bin/python when available.
• Runtime startup filters the known starfile/pkg_resources deprecation warning.

License

This software is licensed under GPL v3.0.

Acknowledgement

Thanks to Jack(Kai) Zhang@Yale MBB for his guidance.

Contributing

Always welcome! This software may still has room for improvement such as updating the usage documentation, improving the GUI design, and enhancing the software's usability.

I am still working on improving this software. More exciting features are on the way!

Contact

If you have any questions, please contact me: zhen.victor.huang@gmail.com