PartiNet 🔬

PartiNet is a three-stage pipeline for automated particle picking in cryo-EM micrographs, combining advanced denoising with state-of-the-art deep learning detection.

Installation and Usage can be found at our Documentation

Use our pretrained model at Model Weights

Features

• 🧹 Heuristic denoising for improved signal-to-noise ratio
• 🎯 Dynamic deep learning particle detection
• ⚡ Multi-GPU support for faster processing
• 🔄 Seamless integration with cryoSPARC and RELION workflows
• 📊 Confidence-based particle filtering
• 🖼️ Visual detection validation

Prerequisites

Before starting, ensure you have:

• Motion-corrected micrographs
• GPU access (recommended)
• PartiNet installation (see Installation section)

Installation

git clone git@github.com:WEHI-ResearchComputing/PartiNet.git
cd PartiNet

Alternatively, use our containers:

# Singularity/Apptainer
apptainer exec --nv --no-home \
    -B /path/to/your/data:/data oras://ghcr.io/wehi-researchcomputing/partinet:main-singularity \
    partinet --help

# Docker
docker pull ghcr.io/wehi-researchcomputing/partinet:main
docker run --gpus all -v /path/to/your/data:/data \
    ghcr.io/wehi-researchcomputing/partinet:main partinet --help

Directory Structure

project_directory/
├── motion_corrected/          # 📁 Input micrographs
├── denoised/                  # 🧹 Denoised outputs
├── exp/                       # 🎯 Detection results
│   ├── labels/               # 📋 Coordinates
│   └── ...                   # 🖼️ Visualizations
└── partinet_particles.star    # ⭐ Final output

Pipeline Stages

1. Denoise

partinet denoise \
  --source /data/my_project/motion_corrected \
  --project /data/my_project

2. Detect

partinet detect \
  --weight /path/to/model_weights.pt \
  --source /data/partinet_picking/denoised \
  --device 0,1,2,3 \
  --project /data/partinet_picking

3. Generate STAR File

partinet star \
  --labels /data/my_project/exp/labels \
  --images /data/my_project/denoised \
  --output /data/my_project/partinet_particles.star \
  --conf 0.1

Output Files

1. Denoised Micrographs (denoised/*.png)

   • Cleaned micrographs with improved SNR

2. Detection Results (exp/)

   • labels/*.txt: Particle coordinates
   • *.png: Visualization overlays

3. STAR File (partinet_particles.star)

   • Ready for RELION processing

Advanced Usage

For detailed information about specific commands:

partinet --help
partinet <command> --help

Available commands:

• denoise: Clean input micrographs
• detect: Identify particles
• star: Generate STAR files
• train: Train custom models

Troubleshooting

• GPU Issues

  • Verify GPU availability: nvidia-smi
  • Check CUDA installation
  • Ensure proper device selection
  • AMD and Intel GPUs are currently untested

• Path Issues

  • Verify directory permissions
  • Check mount points in container setups
  • Ensure absolute paths are used

Contributing

We welcome contributions! Please raise issues or initiate pull requests on this repo.

License

This project is licensed under the terms of MIT license.

Citation

If you use PartiNet in your research, please cite:

Citation information will be added upon publication

Support

For issues and questions:

• Open an Issue
• Check existing Discussions