seudo

Python port (in progress) of SEUDO, a toolbox for removing false transients from calcium imaging source extraction. Original MATLAB implementation and paper: Gauthier & Charles, eLife 2021.

This package is being ported incrementally from the MATLAB codebase (../code in the source repository). Currently implemented:

• The core SEUDO solver (estimate_time_courses_with_seudo) — sparse regression separating each source's true activity from unmodeled ("blob") contamination, for a single frame at a time.
• Transient detection and per-transient spatial profile computation (identify_transients, compute_transient_info).
• Automatic true/false transient classification by profile correlation (auto_classify_transients).
• A PyQt5 GUI (seudo.gui.ClassifyTransientsWindow) for manually reviewing and classifying transients.

Not yet ported: the rwl1df/bpdndf dynamic SEUDO modes, the 'lbq' auto-classification method, parameter search / Bayesian optimization, and several of the original MATLAB GUIs (parameter review, transient detail viewer).

Install

pip install seudo

This includes the PyQt5 classification GUI and h5py (for loading real MATLAB v7.3 .mat files) as standard dependencies, since the GUI is a core part of the workflow, not an add-on.

For local development, from this directory:

pip install -e .[dev]
pytest

Quick start

from seudo import Seudo

# movie: (Y, X, F) array; profiles: (Y, X, nCells) array
se = Seudo(movie, profiles, time_courses=time_courses)

se.compute_transient_info('default')
result = se.estimate_time_courses_with_seudo(sigma2=0.002, lambda_blob=10, blob_radius=3)

To open the manual classification GUI:

from PyQt5 import QtWidgets
from seudo.gui import ClassifyTransientsWindow

app = QtWidgets.QApplication([])
win = ClassifyTransientsWindow(se, 'default')
win.show()
app.exec_()

License

MIT, see LICENSE.