PhenoTypic: A Python Framework for Bio-Image Analysis

A modular image processing framework developed at the NSF Ex-FAB BioFoundry, focused on arrayed colony phenotyping on solid media.

Links:

Overview

PhenoTypic provides a modular toolkit designed to simplify and accelerate the development of reusable bio-image analysis pipelines. PhenoTypic provides bio-image analysis tools built-in, but has a streamlined development method to integrate new tools.

Installation

uv (recommended)

See more on installing uv

Regular Install (recommended when deploying on a cluster)

uv add phenotypic

Interactive / GUI Install (Plotly dashboards, Jupyter, Dash hub)

uv add phenotypic --extra gui

napari Desktop Viewer Install (for image.*.napari(), point picker, sweep viewer)

uv add phenotypic --extra napari

Pip

Regular Install

pip install phenotypic

Interactive / GUI Install

pip install "phenotypic[gui]"

napari Desktop Viewer Install

pip install "phenotypic[napari]"

Note: may not always be the latest version. Install from repo when latest update is needed

Manual Installation (For latest updates)

git clone https://github.com/exfab/PhenoTypic.git
cd PhenoTypic
uv sync

Dev Installation

For extending PhenoTypic.

git clone https://github.com/exfab/PhenoTypic.git
cd PhenoTypic
uv sync --group dev --all-extras

CPU compatibility (polars build)

PhenoTypic installs the polars-lts-cpu build by default so it runs on older CPUs (e.g. pre-AVX2 HPCC nodes) without crashing. On AVX2-capable machines you can swap in the faster stock polars build for quicker measurement compilation — see Choosing the polars CPU build.

GPU-Accelerated Detection (SAM2, micro-sam)

PhenoTypic ships optional deep-learning detectors backed by Meta's Segment Anything Model 2 and micro-sam.

See GPU Detection Setup for model downloads and SLURM deployment instructions.

Optional third-party tools

libvips (faster GUI deep-zoom preparation)

On macOS and Windows, the [gui] extra installs the official pyvips[binary] distribution, including a self-contained native libvips. No Homebrew install or loader-path configuration is needed for the normal GUI installation.

Linux and HPC installations keep the smaller pyvips binding and may provide native libvips through the operating system or an environment module:

# Debian or Ubuntu
sudo apt install libvips-dev --no-install-recommends

Verify that Python can load the native library from the same shell that will launch the GUI:

uv run python -c "import pyvips; print('pyvips', pyvips.__version__, 'libvips', '.'.join(str(pyvips.version(i)) for i in range(3)))"

The bundled build contains the common image loaders needed by the GUI but omits optional facilities such as PDF and OpenSlide. Users who intentionally prefer a fuller Homebrew libvips can install it with brew install vips. If vips --version then works but Python cannot find libvips.42.dylib, expose Homebrew's library directory before launching PhenoTypic:

export DYLD_FALLBACK_LIBRARY_PATH="$(brew --prefix vips)/lib${DYLD_FALLBACK_LIBRARY_PATH:+:$DYLD_FALLBACK_LIBRARY_PATH}"

If native libvips or its loader is unavailable on any platform, PhenoTypic automatically retains the Pillow fallback. Browse still works, but preparing large pyramids can be slower and use more memory.

ExifTool (RAW metadata)

To extract metadata from raw images, PhenoTypic uses the PyExifTool module. This requires the external ExifTool application. If it is unavailable, some RAW metadata may not be imported. See the PyExifTool dependency documentation.

Chrome or Chromium (PNG export of Plotly figures)

Plotly figures are rasterised by Kaleido, which drives a Chrome or Chromium browser. The browser is not installed by uv sync. Without it:

• Pipeline plots (ImagePipeline(plots=[...])) still publish: Plotly plots write interactive .html pages and skip the .png. Matplotlib plots are unaffected. CLI validation logs one warning naming the affected plots before the run starts.
• The results viewer's Scatter-tab PDF export reports the missing browser beside its button instead of exporting.

Install a browser Kaleido can use:

uv run plotly_get_chrome

or point the BROWSER_PATH environment variable at a Chrome/Chromium binary you already have. See Custom pipeline plot for what each backend publishes.

Run the CLI

Process a directory of plate images through a saved pipeline:

uv run python -m phenotypic --mode full --pipeline pipeline.json --input ./images --output ./out

Use --mode process --layer {rgb|gray|detect_mat|objmap} for an apply-only export run that writes a single image layer per input (mirroring the input tree) and skips the measurement/analysis suite — handy for previewing detection or enhanced layers.

Launch the GUI

The unified GUI hub bundles the pipeline builder, results viewer, and run console under one URL. Start it with the phenotypic-gui console script:

uv run phenotypic-gui --root ./images --port 8050

--root freezes the sandbox the GUI's file browser is allowed to see (defaults to the current working directory). --host 127.0.0.1 (the default) keeps the server loopback-only — pair with SSH port forwarding for remote workstations:

ssh -L 8050:localhost:8050 user@cluster

Open http://localhost:8050/ in your browser. The GUI hub guide walks through the file browser, builder, run console, and results viewer.

For Open OnDemand-style proxies, pass only the browser-visible path prefix:

uv run phenotypic-gui --root /rhome/ejaco020 --host 0.0.0.0 --port 30099 --url-prefix /node/hz01/30099/

Then open the full proxy URL, for example https://ondemand.hpcc.ucr.edu/node/hz01/30099/.

Note: phenotypic gui (no hyphen, as a subcommand) is not supported. Use phenotypic-gui.

Hyperparameter Tuning

Search an ImagePipeline's parameters to minimize a scorer-defined cost with the tuning engine:

uv run phenotypic-tune run spec.json -i ./plates -o ./out

Grid and random search work out of the box; the Optuna samplers (tpe/cmaes/gp/nsga2) need the tune extra. See the tuning how-to for an end-to-end walkthrough.

Module Overview

Sponsors