Easymode is a collection of general pretrained neural networks for cellular cryo-electron tomography (cryoET). It provides single command line interface functions to handle feature detection: segmentation of cellular features, coordinate extraction, and data preprocessing. Inspired by, built on, and built to work with: Warp/M, RELION, MemBrain-seg, Ais, and Pom.

Preprint | Documentation | Model library | Tutorials

conda create -n easymode python=3.10 cudatoolkit=11.2 cudnn=8.1 git -c conda-forge
conda activate easymode
pip install tensorflow==2.11.0
pip install git+https://github.com/mgflast/easymode.git

For full installation instructions including CUDA setup, Ais, and Pom, see the installation guide.

# List available models
easymode list

# Segment features
easymode segment ribosome microtubule --data warp_tiltseries/reconstruction

# Extract particle coordinates
easymode pick ribosome --data segmented/ --output coordinates/ribosome --size 2000000 --spacing 300

# Reconstruct tomograms (requires WarpTools and AreTomo3)
easymode reconstruct --frames frames/ --mdocs mdocs/ --apix 1.56 --dose 4.6

# Denoise tomograms
easymode denoise --method n2n --mode direct --data warp_tiltseries/reconstruction --output denoised

Pretrained networks are hosted on HuggingFace and downloaded automatically on first use. See the user guide for detailed usage.

All networks were trained on a dataset of over 4500 tilt series from 68+ sources, covering many prokaryotic, archaeal, and eukaryotic species, different sample preparation techniques, hardware configurations, pixel sizes, electron doses, and defocus values. Tomograms were reconstructed in multiple flavours using WarpTools, AreTomo3, cryoCARE, and DeepDeWedge at 10.0 A voxel size. See the preprint for more details.

Mart So-Last, 2026 | mgflast@gmail.com