torch-tilt-series

Tilt series data structure, projection and subtilt extraction for cryo-ET.

Overview

This package provides a TiltSeries class for working with cryo-ET tilt series alignment geometry in PyTorch, plus loading and preprocessing for the raw tilt images it describes. It supports

• loading alignment metadata from AreTomo (.aln) and ETOMO directories using alnfile and etomofiles
• storing tilt series alignment parameters (tilt angles, tilt axis angles, translations, x-tilts) and coordinate-space transforms (see the TiltSeries docstring)
• computing projection matrices and projecting 3D points into 2D detector coordinates
• load_tilt_series_images(): loading the raw tilt images a TiltSeries points to (image_path/image_indices), via mrcfile
• normalize_on_central_crop() / subtract_plane(): normalization and linear background subtraction building blocks (torch_tilt_series.utils)
• preprocess_tilt_series_images(): a full per-image preprocessing pipeline - background plane subtraction, a bandpass filter (mirror-padded and soft-masked to suppress edge artifacts), and central-crop normalization (torch_tilt_series.preprocessing)

All 3D/2D positions are in zyx/yx coordinates, in Angstroms, relative to the tomogram/detector center. Translations are stored in Angstroms as (y, x).

Subtilt/subvolume extraction and full volume reconstruction live in torch-reconstruct-tomogram.

Installation

pip install torch-tilt-series

To load alignment data from AreTomo or ETOMO files, or raw tilt images via load_tilt_series_images(), install the optional IO dependencies:

pip install torch-tilt-series[io]

Examples

See the examples/ folder for scripts showing how to load a tilt series and use the API.

License

This project is licensed under the BSD 3-Clause License - see the LICENSE file for details.