Spotiflow is a deep learning-based, threshold-agnostic, subpixel-accurate 2D and 3D spot detection method for fluorescence microscopy. It is primarily developed for spatial transcriptomics workflows that require transcript detection in large, multiplexed FISH-images, although it can also be used to detect spot-like structures in general fluorescence microscopy images and volumes. A more detailed description of the method can be found in our paper.
The documentation of the software can be found here.
Create and activate a fresh conda environment (we currently support Python 3.9 to 3.12):
conda create -n spotiflow python=3.12
conda activate spotiflowNote (for MacOS users): if using MacOS, there is a known bug causing the installation of PyTorch with conda to sometimes break OpenMP. You can avoid installing PyTorch with conda and let spotiflow install it automatically via pip instead.
For Linux/Windows with a CUDA device, install PyTorch using conda/mamba (one might need to change the cuda version accordingly):
conda install pytorch torchvision pytorch-cuda=12.1 -c pytorch -c nvidia # Might need to change the cuda version accordinglyNote (for Windows users): if using Windows, please install the latest Build Tools for Visual Studio (make sure to select the C++ build tools during installation) before proceeding to install Spotiflow.
Finally, install spotiflow:
pip install spotiflowFor Linux/MacOS users, you can also install Spotiflow using conda through the conda-forge channel:
conda install -c conda-forge spotiflowThe conda package is, for now, not CUDA-compatible. We recommend using pip to install Spotiflow if available.
The CLI is the easiest way to train (or fine-tune) a model. To train a model, you can use the following command:
spotiflow-train INPUT_DIR -o OUTPUT_DIRwhere INPUT_DIR is the path to the directory containing the data in the format described here and OUTPUT_DIR is the directory where the trained model will be saved. You can also pass other parameters to the training, such as the number of epochs, the learning rate, etc. For more information, including examples, please refer to the training documentation or run the command spotiflow-train --help.
For training with the API, please check the training example notebook. For finetuning an already pretrained model, please refer to the finetuning example notebook.
3D models can also be trained with the CLI by adding the --is-3d True flag, as shown below:
spotiflow-train INPUT_DIR -o OUTPUT_DIR --3d TrueSee the example 3D training script for an API example. For more information, please refer to the 3D training example notebook. Fine-tuning a 3D model can be done by following the same workflow as to the 2D case.
You can use the CLI to run inference on an image or folder containing several images. To do that, you can use the following command:
spotiflow-predict PATHwhere PATH can be either an image or a folder. By default, the command will use the general pretrained model. You can specify a different model by using the --pretrained-model flag. Moreover, spots are saved to a subfolder spotiflow_results created inside the input folder (this can be changed with the --out-dir flag). For more information, please refer to the help message of the CLI ($ spotiflow-predict -h).
Alternatively to installing Spotiflow as command line tool on your operating system, you can also use it directly from our Docker container (thanks to @migueLib for the contribution!). To do so, you can use the following command:
To pull the Docker container from Dockerhub use:
docker pull weigertlab/spotiflow:mainThen, run spotiflow-predict with:
docker run -it -v [/local/input/folder]:/spotiflow/input weigertlab/spotiflow:main spotiflow-predict input/your_file.tif -o .Where:
-v: represents the volume flag, which allows you to mount a folder from your local machine to the container.
/path/to/your/data:/spotiflow: is the path to the folder containing the image you want to analyze.
Note:
- The current implementation of Spotiflow in Docker only supports CPU inference.
The API allows detecting spots in a new image in a few lines of code! Please check the corresponding example notebook and the documentation for a more in-depth explanation. The same procedure can be followed for 3D volumes.
from spotiflow.model import Spotiflow
from spotiflow.sample_data import test_image_hybiss_2d
# Load sample image
img = test_image_hybiss_2d()
# Or any other image
# img = tifffile.imread("myimage.tif")
# Load a pretrained model
model = Spotiflow.from_pretrained("general")
# Or load your own trained model from folder
# model = Spotiflow.from_folder("./mymodel")
# Predict
points, details = model.predict(img) # points contains the coordinates of the detected spots, the attributes 'heatmap' and 'flow' of `details` contain the predicted full resolution heatmap and the prediction of the stereographic flow respectively (access them by `details.heatmap` or `details.flow`). Retrieved spot intensities are found in `details.intens`.Our napari plugin allows detecting spots in 2D and 3D directly with an easy-to-use UI. See napari-spotiflow for more information.
We provide several pre-trained models that may be used out-of-the-box. The available models are: general, hybiss, synth_complex, synth_3d and smfish_3d. For more information on these pre-trained models, please refer to the article and the documentation.
The default cache directory root folder (where pre-trained models and datasets are stored) is, by default, ~/.spotiflow. If you want to change it for your use case, you can either set the environment variable SPOTIFLOW_CACHE_DIR to the path you want or directly pass the desired folder as an argument (cache_dir) to the Spotiflow.from_pretrained() method (note that if the latter is chosen, the path stored in the environment variable will be ignored).
Spotiflow can be seamlessly integrated in existing Starfish pipelines using our spotiflow.starfish.SpotiflowDetector as a spot detection method instead of the BlobDetection classes shipped with Starfish, requiring minimal code changes apart from the addition of Spotiflow to the existing environment where Starfish is installed. For an example, please refer to the provided script.
We are open to contributions, and we indeed very much encourage them! Make sure that existing tests pass before submitting a PR, as well as adding new tests/updating the documentation accordingly for new features.
First, clone the repository:
git clone [email protected]:weigertlab/spotiflow.gitThen install the testing extras:
cd spotiflow
pip install -e ".[testing]"then run the tests:
pytest -v --color=yes --cov=spotiflowInstall the docs extras:
pip install -e ".[docs]"and then cd into the docs folder of the cloned repository and build them:
cd spotiflow/docs
sphinx-build -M html source buildIf you use this code in your research, please cite the Spotiflow paper (currently preprint):
@article {dominguezmantes24,
author = {Dominguez Mantes, Albert and Herrera, Antonio and Khven, Irina and Schlaeppi, Anjalie and Kyriacou, Eftychia and Tsissios, Georgios and Skoufa, Evangelia and Santangeli, Luca and Buglakova, Elena and Durmus, Emine Berna and Manley, Suliana and Kreshuk, Anna and Arendt, Detlev and Aztekin, Can and Lingner, Joachim and La Manno, Gioele and Weigert, Martin},
title = {Spotiflow: accurate and efficient spot detection for fluorescence microscopy with deep stereographic flow regression},
elocation-id = {2024.02.01.578426},
year = {2024},
doi = {10.1101/2024.02.01.578426},
publisher = {Cold Spring Harbor Laboratory},
URL = {https://www.biorxiv.org/content/early/2024/02/05/2024.02.01.578426},
eprint = {https://www.biorxiv.org/content/early/2024/02/05/2024.02.01.578426.full.pdf},
journal = {bioRxiv}
}