Skip to content

Workflows

CryoSiam workflows are designed to process cryo-electron tomography (CryoET) data step by step. Each module focuses on a specific task, and you can combine them into analysis pipelines depending on your scientific question.

For details, please refer to the CryoSiam preprint.

Workflow 1: Denoising → Semantic Segmentation / Particle Identification

  1. Denoising
    Clean the raw tomogram to reduce noise while preserving structural details.
  2. Semantic Segmentation
    Classify each voxel into biological classes such as membranes, filaments, or complexes.
    Alternative: Particle Identification can be used to locate specific particles of interest directly after denoising.

See details in Denoising, Semantic Segmentation, and Particle Identification.

Workflow 2: Denoising → Instance Segmentation

  1. Denoising
    Prepare a cleaner tomogram for reliable downstream processing.
  2. Instance Segmentation
    Separate individual structures even when they overlap or belong to the same class.

See details in Denoising and Instance Segmentation.

Workflow 3: Denoising → Instance Segmentation → Subtomogram Embeddings

  1. Denoising
    Preprocess tomograms for structural clarity.
  2. Instance Segmentation
    Extract and separate candidate subtomograms.
  3. Subtomogram Embeddings
    Represent subtomograms as feature vectors for clustering, comparison, or downstream analysis.

See details in Denoising, Instance Segmentation, and Subtomogram Embeddings.

Trained Models

The available trained models are described here: Trained Models.

Configuration Files

Each module requires a YAML configuration file defining inputs, outputs, and model parameters.
You can run a module as:

cryosiam <module> --config_file=configs/<module>.yaml

Explanation of the YAML configuration files is given into the specific documentation page.