Raw and annotated data for Shape2Fate

Datasets acquired and generated for the manuscript "Shape2Fate: a morphology-aware deep learning framework for tracking endocytic and exocytic carriers at nanoscale". The datasets include test, training and time series datasets each containing the raw data and the annotated data where it applies. 

Overview

This archive contains live-cell time-lapse TIRF-SIM datasets used in the development of Shape2Fate. Each dataset includes raw SIM data (all phases/orientations/frames in a single file). Depending on the experiment, additional files include SIM reconstructions (-recon), binary masks (-mask.tif), and manual annotations (-annotations.csv).  
Unless noted otherwise, data is provided in .dv format.

This dataset is related to the following paper:
Shape2Fate: a morphology-aware deep learning framework for tracking endocytic and exocytic carriers in super-resolution TIRF-SIM. 

Biological systems

Datasets were acquired from engineered mammalian cell lines expressing fluorescent reporters of clathrin-mediated endocytosis and cargo exocytosis. The following reporters are represented: EGFP-CLCa, SNAP-CLCa (JF-dye labelled), SBP-mEmerald-LAMP1, pHluorin-GLUT4, and Dynamin2-mRuby3.  
RUSH assays (_STIM) were stimulated with biotin; adipocyte GLUT4 datasets (_STIM) were stimulated with insulin. Further methodological details are provided in the associated manuscript.

Imaging

Most TIRF-SIM experiments were carried out at the Micron Bioimaging Facility, University of Oxford, using a DeltaVision OMX SR inverted wide-field microscope equipped with a 60× 1.5-NA TIRF UPLAPO oil-immersion objective (Olympus) and a PCO Edge 4.2 sCMOS camera. Acquisition parameters are stored in the raw files metadata.

Channel registration

Multicolour datasets were registered using 0.1 µm TetraSpeck™ fluorescent microspheres imaged under identical conditions as the experimental datasets. Registration was performed using the sub-pixel alignment method of Thevenaz et al. (IEEE Trans. Image Process., 1998; doi: 10.1109/83.650848), applying a similarity (scaled rotation) transformation. For each dataset, a corresponding transformation matrix is provided (registration_transform.json), mapping coordinates from the target channel to the source channel (i.e., inverse transform).

Masks

Each image file is accompanied by a corresponding binary mask that delineates the in-focus, well-adhered regions of cells within the field of view. These masks were used to restrict all quantitative analyses to areas containing reliably imaged cells.

Dataset groups  

Below is a summary of the included datasets:

The exocytosis productivity annotations in the Adipocytes-CME coupling and the RUSH-CME local coupling datasets are split at the cell level into training and validation subsets.

• Adipocytes-CME coupling - cells acquired on 2025/02/25 were assigned to the training split (cell index < 240), and cells acquired on 2025/02/26 were assigned to the validation split (cell index >= 240).
• RUSH-CME local coupling - cells acquired before 5pm were assigned to the training split (cell index < 162); cells acquired after 5pm were assigned to the validation split (cell index >= 162).