Interpretable super-resolution dimension reduction of spatial transcriptomics data by DeepFuseNMF
Project description
Interpretable super-resolution dimension reduction of spatial transcriptomics data by DeepFuseNMF
Overview
DeepFuseNMF is based on a multi-modal neural network that takes advantage of the high-dimensionality of transcriptomics data and the super-resolution of image data to achieve interpretable super-resolution dimension reduction. The high-dimensional expression data enable refined functional annotations and the super-resolution image data help to enhance the spatial resolution.
Based on the super-resolution embedding and the reconstruction of gene expressions, DeepFuseNMF can then perform super-resolution downstream analyses, such as spatial domain detection, gene expression recovery, and identification of embedding-associated genes as well as super-resolution cluster-associated genes.
Installation
Please install DeepFuseNMF
from pypi with:
pip install deepfusenmf
Or clone this repository and use
pip install -e .
in the root of this repository.
Quick start
Prepare your data and run the following command:
python run_DeepFuseNMF.py --config config.json --device 0 --verbose
config.json
is a JSON file that contains the paths to the input data and the output directory. The JSON file should look like this:
{
"settings": {
"root_path": "your_root_path",
"project": "your_project_name"
},
"sections": [
{
"name": "name of section A",
"image_path": "image path of section A",
"spot_coord_path": "spot coordinate path of section A",
"spot_exp_path": "spot expression path of section A"
},
{ // multi-sections is supported
"name": "name of section B",
"image_path": "image path of section B",
"spot_coord_path": "spot coordinate path of section B",
"spot_exp_path": "spot expression path of section B"
}
],
"paras": {
"scale_factor": [2, 2], // the scale factor of the input images, integer for single section, list for multi-sections
"radius": [65, 65] // the radius of the spots, integer for single section, list for multi-sections
}
}
All parameters of run_DeepFuseNMF.py
are optional. The following are the:
--config
or-c
: the path to the configuration file--rank
or-r
: the rank / number of components of the NMF model, default is 20--seed
or-s
: the random seed, default is 123--device
or-d
: the device to run the model, e.g., 0, 1, 2, etc, default is 0--save_score
: whether to save the model, default isTrue
--spot_wise
: whether to save the spot-wise results, use--spot_wise
to save the spot-wise results--verbose
: whether to print the log information, use--verbose
to print the log information
Please refer to the Tutorial/Tutorial.ipynb
and configs/README.md
for more details.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for DeepFuseNMF-0.0.1-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 52d47fc79af1c335fc51d2522a7e15777bf1bbb36c7a04ca6e0e007d3f60267a |
|
MD5 | 19e122c721ed0eca0170b1f6fd25b577 |
|
BLAKE2b-256 | 9ba31fa376be32df54098e3c5e91aeec2b3a0b9a0be3b132df68ba22126404bc |