Code and example imaging data for "Resolving single-cell secretion dynamics underlying cell–cell communication"

Computational decoding framework for time-resolved FluoroSpot, a live-cell secretion imaging platform. In a sandwich immunoassay the fluorescence recorded at each pixel is the cumulative amount of detection antibody bound over time, so secretion dynamics are blurred by signal accumulation and antibody-binding kinetics. This software inverts that encoding — a spatial and temporal Gaussian, a Savitzky–Golay temporal derivative, and a Fourier deconvolution of the antibody-binding impulse response — to reconstruct instantaneous single-cell secretion rates from the accumulated readout.

This deposit accompanies the manuscript “Resolving single-cell secretion dynamics underlying cell–cell communication” (Yang et al.). It contains the source code and the representative time-lapse datasets published as Fig. 2e, Fig. 4b and Fig. 5a. Each dataset is the published field of view, cut from the drift-corrected full-frame recording with a margin around it so that the published region is unaffected by the spatial filtering. Decoding these crops reproduces the published reconstruction.

How to use this deposit

Download all files. Unpack the code archive, then move the ten data files into the data/ directory inside it — the notebooks expect them there:

trfluorospot-1.0.0/
├── notebooks/          decode_images.ipynb, supplementary_figures.ipynb
├── data/               <-- put the ten downloaded .tif / .csv files here
└── environment_cpu.yml, environment_gpu.yml

Then create the environment (mamba env create -f environment_cpu.yml or environment_gpu.yml for CUDA), pip install -e ., and run the two notebooks from the notebooks/ directory. A GPU is optional. See README.md for the tested configuration and data/README.md for checksums and the provenance of every crop.

Contents

trfluorospot-1.0.0.zip — source code and notebooks (see README.md)

Fig. 2e — mouse alveolar macrophages + silica nanoparticles, IL-1β
fig2e_il1b.tif — 732 × 258 × 308, uint16 (decoder input)

Fig. 4b — mouse alveolar macrophages + nigericin, IL-1β
fig4b_il1b.tif — 610 × 366 × 366, uint16 (decoder input)
fig4b_dic.tif — DIC channel, 610 × 366 × 366, uint16
fig4b_sytox.tif — SYTOX Orange, 610 × 366 × 366, uint16

Fig. 5a — a cytotoxic T lymphocyte attacking a YTN16 cancer cell, granzyme B. All image stacks are 2128 × 239 × 239, uint16.
fig5a_granzymeb.tif — decoder input
fig5a_dic.tif — DIC channel
fig5a_cellbrite.tif — CellBrite Orange membrane label
fig5a_yopro.tif — YO-PRO-1 cell-death channel
fig5a_mask.tif — cancer-cell label
fig5a_tracks.csv — trajectory of the attacking lymphocyte

The complete raw time-lapse recordings are 0.1–1.6 TB per experiment (about 3.7 TB across the study) and are available from the corresponding author on reasonable request.