Partaker

Partaker (Python-based Analyzer for Real-Time Assessment of Kinetics and Expression in Real-time) is a GUI application for single-cell bacterial image analysis in microfluidic time-lapse microscopy. It integrates deep learning segmentation, multi-channel fluorescence quantification with Relative Promoter Unit (RPU) calibration, and morphological analysis in a unified interface.

Partaker allows you to create plots like this:

Features

• Multi-model segmentation: Six selectable deep learning models for cell segmentation
• Multi-channel fluorescence: Per-cell fluorescence quantification with background subtraction and RPU calibration
• Morphological analysis: Automated cell classification (rod, coccoid, artifact) with eight morphological descriptors
• ND2 and TIFF support: Native reading of Nikon ND2 files and standard TIFF stacks
• Multi-file stitching: Automatic concatenation of sequential acquisitions on the time axis
• Session management: Save and resume analysis sessions through HDF5

Segmentation Models

Partaker integrates the following pretrained deep learning backends:

Model	Family	Best For
omnipose_bact_phase	Omnipose	Phase-contrast, dense monolayers
omnipose_bact_fluo	Omnipose	Fluorescence images
bact_phase_cp3	Cellpose 3	Phase-contrast, general bacteria
bact_fluor_cp3	Cellpose 3	Fluorescence images
cellpose_deepbacs	DeepBacs/Cellpose	Phase-contrast bacteria
unet	Custom U-Net	User-trained binary segmentation

Users can compare model outputs on their data and select the best-performing model for their imaging conditions.

Requirements

• Python >= 3.10, < 3.12
• macOS (Apple Silicon or Intel), Windows, or Linux
• uv package manager (recommended)

Installation

1. Install uv if you do not have it:

curl -LsSf https://astral.sh/uv/install.sh | sh

2. Clone the repository:

git clone https://github.com/SamOliveiraLab/partaker.git
cd partaker

3. Launch Partaker:

uv run gui

This will automatically download and install all dependencies and open the main GUI window.

Quick Start

1. Load data: Go to File > Open and select your ND2 or TIFF file
2. Navigate: Use the T (time), P (position), and C (channel) sliders to browse your data
3. Segment: Select a segmentation model from the dropdown (e.g., omnipose_bact_phase) and switch Display Mode to "Labeled Segmentation" or "Overlay with Outlines"
4. Batch segment: Use the Segmentation tab on the right panel to select positions, time range, and model, then click "Segment Selected"
5. Analyze fluorescence: Switch to the Population tab to quantify fluorescence intensity per cell across channels
6. Morphology: Use the Morphology tab to extract cell shape descriptors and classify cell types

Display Modes

• Normal: Raw microscopy image
• Overlay with Outlines: Segmentation contours overlaid on the original image
• Labeled Segmentation: Color-coded instance labels showing individual cells

Custom U-Net Weights

To use a custom-trained U-Net model, set the environment variable before launching:

export PARTAKER_UNET_WEIGHTS="/path/to/your/unet_weights.pt"
uv run gui

The weights file should be a PyTorch state-dict (.pt). A conversion script (convert.py) is provided for converting Keras .h5 weights to PyTorch format.

Example Dataset

The validation dataset used in the manuscript (two-strain E. coli co-culture in monolayer microfluidic chambers) is available on the BioImage Archive under accession S-BIAD3015.

Dataset

The segmentation benchmark dataset (45 annotated phase-contrast frames with ground-truth instance masks across four chamber positions) and the custom U-Net training dataset (80 frames with binary and instance masks, boundary weight maps, and trained model weights) are archived on Zenodo, together with the benchmarking notebook used to compute the reported metrics:

https://doi.org/10.5281/zenodo.20577330

Expected Outputs

Running the segmentation and analysis workflow on the example data produces:

• Segmentation masks: instance-labeled masks for each frame, viewable as "Labeled Segmentation" or "Overlay with Outlines"
• Cell metrics CSV: a per-cell, per-frame table containing morphological descriptors (area, perimeter, aspect ratio, solidity, and others) and per-channel fluorescence values
• Fluorescence time series: aggregated single-cell and population-level RPU traces synchronized with medium perturbations
• Benchmark metrics CSV: per-frame instance-level Jaccard Index, Dice coefficient, and mean IoU for each segmentation model (see the benchmarking notebook in the Zenodo deposit)

Demo

A video walkthrough of Partaker is available on YouTube: Partaker Demo

Documentation

Full documentation is available at: https://samoliveiralab.github.io/partaker/

Contributing

Contributions are welcome. Please use conventional commits and contact the maintainers before starting major changes.