| Name | step-kit JSON |
| Version |
0.2.6
JSON |
| download |
| home_page | https://github.com/SGGb0nd/step |
| Summary | STEP, an acronym for Spatial Transcriptomics Embedding Procedure, is a deep learning-based tool for the analysis of single-cell RNA (scRNA-seq) and spatially resolved transcriptomics (SRT) data. STEP introduces a unified approach to process and analyze multiple samples of scRNA-seq data as well as align several sections of SRT data, disregarding location relationships. Furthermore, STEP conducts integrative analysis across different modalities like scRNA-seq and SRT. |
| upload_time | 2024-11-05 05:13:49 |
| maintainer | None |
| docs_url | None |
| author | SGGb0nd |
| requires_python | <4.0,>=3.10 |
| license | Apache-2.0 |
| keywords |
spatial transcriptomics
single-cell rna-seq
deep learning
scrna-seq
srt
step
step
|
| VCS |
 |
| bugtrack_url |
|
| requirements |
No requirements were recorded.
|
| Travis-CI |
No Travis.
|
| coveralls test coverage |
No coveralls.
|
# STEP: Spatial Transcriptomics Embedding Procedure
[](https://github.com/SGGb0nd/step/actions/workflows/mkdocs.yaml)
[](https://github.com/SGGb0nd/step/actions/workflows/pages/pages-build-deployment)
[](https://badge.fury.io/py/step-kit)
[](https://doi.org/10.1101/2024.04.15.589470)
## Introduction
STEP, an acronym for Spatial Transcriptomics Embedding Procedure, is a foundation deep learning/AI architecture for the analysis of spatially resolved transcriptomics (SRT) data, and is also compatible with scRNA-seq data. STEP roots on the precise captures of three major varitions occured in the SRT (and scRNA-seq) data: **Transcriptional Variations**, **Batch Variations** and **Spatial Variations** with the correponding modular designs: **Backbone model**: a Transformer based model togther with gene module seqeunce mapping; **Batch-effect model**: A pair of inverse transformations utilizing the *batch-embedding* conception for the decoupled batch-effect elimination; **Spatial model**: a GCN-based spatial filter/smoother working on the extracted embedding from the Backbone model, different from the usage of GCN in other methods as a feature extractor. Thus, with the proper combinations of these models, STEP introduces a unified approach to systematically process and analyze single or multiple samples of SRT data, disregarding location relationships between sections (meaning both contiguous and non-contiguous sections), to reveal multi-scale bilogical heterogeneities (cell types and spatial domains) in multi-resolution SRT data. Furthermore, STEP can also conduct integrative analysis on scRNA-seq and SRT data.
## Key Features
- Integration of multiple scRNA-seq and single-cell resolution SRT samples to reveal cell-type level heterogeneities.
- Alignment of various SRT data sections contiguous or non-contiguous to identify spatial domains across sections.
- Scalable to different data resolutions, i.e., wild range of technologies and platforms of SRT data, including **Visium HD**, **Visum**, **MERFISH**, **STARmap**, **Stereo-seq**, **ST**, etc.
- Scalable to large datasets with a high number of cells and spatial locations.
- Performance of integrative analysis across modalities (scRNA-seq and SRT) and cell-type deconvolution for the non-single-cell resolution SRT data.
## Other Capabilities
- Capability to produce not only the batch-corrected embeddings but also batch-corrected gene expression profiles for scRNA-seq data.
- Capability to perform spatial mapping of reference scRNA-seq data points to the spatial locations of SRT data based on the learned co-embeddings and kNN.
- Comprehensive `adata` processing by specifically desinged `BaseDataset` class and its view version `MaskedDataset` class for the SRT data, which can be easily integrated with the PyTorch `DataLoader` class for training and validation.
- Low computational cost and high efficiency in processing large-scale SRT data: 4-8 GB GPU memory is sufficient for processing a dataset with 100,000 spatial locations with 2000+ sample-size in fast mode, and consumes less than 6 minutes for training in 2000 iterations (tested on NVIDIA RTX 3090).
## System Requiremtns
- **Software Requirements**: Python 3.10+, CUDA 11.6+, PyTorch 1.13.1+, and dgl 1.1.3+.
- **Hardware Requirements**: NVIDIA GPU with CUDA support (recommended), 8GB+ GPU memory. As possible as high RAM and CPU cores for storing and processing large-scale data (this is not required by STEP, but by the data itself).
## Installation
### Quick Installation (Recommended for Python 3.10, Linux, CUDA 11.7)
```bash
pip install step-kit[cu117]
```
This command installs STEP along with compatible versions of PyTorch and DGL.
### Basic Installation
```bash
pip install step-kit
```
⚠️ **Critical Note**:
- This basic installation **does not** include PyTorch and DGL.
- STEP **will not function** without PyTorch and DGL properly installed.
- You must install these dependencies separately before using STEP.
### Manual Dependency Installation
If you need to install specific versions of PyTorch and DGL:
1. Install DGL (example for latest DGL compatible with PyTorch 2.3.x and CUDA 11.8):
```bash
pip install dgl -f https://data.dgl.ai/wheels/torch-2.3/cu118/repo.html
```
2. Install PyTorch (example for PyTorch 2.3.1 with CUDA 11.8):
```bash
pip install torch==2.3.1+cu118 -f https://download.pytorch.org/whl/cu118/torch_stable.html
```
For more information on installing PyTorch and DGL, visit:
- PyTorch: https://pytorch.org/
- DGL: https://www.dgl.ai/
Remember to choose versions compatible with your system configuration and CUDA version (if applicable).
## Contribution
We welcome contributions! Please see [`CONTRIBUTING.md`](./CONTRIBUTING.md) for more details!
## License
step is licensed under [LICENSE](./LICENSE)
## Contact
If you have any questions, please feel free to contact us at [here](mailto:lilounan1997@gmail.com), or feel free to open an issue on this repository.
## Citation
The preprint of STEP is available at [bioRxiv](https://www.biorxiv.org/content/early/2024/04/20/2024.04.15.589470.full.pdf). If you use STEP in your research, please cite:
```bibtex
@article{Li2024.04.15.589470,
title = {{{STEP}}: {{Spatial}} Transcriptomics Embedding Procedure for Multi-Scale Biological Heterogeneities Revelation in Multiple Samples},
author = {Li, Lounan and Li, Zhong and Li, Yuanyuan and Yin, Xiao-ming and Xu, Xiaojiang},
year = {2024},
journal = {bioRxiv : the preprint server for biology},
eprint = {https://www.biorxiv.org/content/early/2024/04/20/2024.04.15.589470.full.pdf},
publisher = {Cold Spring Harbor Laboratory},
doi = {10.1101/2024.04.15.589470},
}
```
Raw data
{
"_id": null,
"home_page": "https://github.com/SGGb0nd/step",
"name": "step-kit",
"maintainer": null,
"docs_url": null,
"requires_python": "<4.0,>=3.10",
"maintainer_email": null,
"keywords": "spatial transcriptomics, single-cell RNA-seq, deep learning, scRNA-seq, SRT, step, STEP",
"author": "SGGb0nd",
"author_email": "lilounan1997@gmail.com",
"download_url": "https://files.pythonhosted.org/packages/a4/1a/59da27dcb04290f526a2f5e07ce9d5122515456f369268c46c01b292da11/step_kit-0.2.6.tar.gz",
"platform": null,
"description": "# STEP: Spatial Transcriptomics Embedding Procedure\n[](https://github.com/SGGb0nd/step/actions/workflows/mkdocs.yaml)\n[](https://github.com/SGGb0nd/step/actions/workflows/pages/pages-build-deployment)\n[](https://badge.fury.io/py/step-kit)\n[](https://doi.org/10.1101/2024.04.15.589470)\n\n\n\n\n## Introduction\nSTEP, an acronym for Spatial Transcriptomics Embedding Procedure, is a foundation deep learning/AI architecture for the analysis of spatially resolved transcriptomics (SRT) data, and is also compatible with scRNA-seq data. STEP roots on the precise captures of three major varitions occured in the SRT (and scRNA-seq) data: **Transcriptional Variations**, **Batch Variations** and **Spatial Variations** with the correponding modular designs: **Backbone model**: a Transformer based model togther with gene module seqeunce mapping; **Batch-effect model**: A pair of inverse transformations utilizing the *batch-embedding* conception for the decoupled batch-effect elimination; **Spatial model**: a GCN-based spatial filter/smoother working on the extracted embedding from the Backbone model, different from the usage of GCN in other methods as a feature extractor. Thus, with the proper combinations of these models, STEP introduces a unified approach to systematically process and analyze single or multiple samples of SRT data, disregarding location relationships between sections (meaning both contiguous and non-contiguous sections), to reveal multi-scale bilogical heterogeneities (cell types and spatial domains) in multi-resolution SRT data. Furthermore, STEP can also conduct integrative analysis on scRNA-seq and SRT data.\n\n## Key Features\n\n- Integration of multiple scRNA-seq and single-cell resolution SRT samples to reveal cell-type level heterogeneities.\n- Alignment of various SRT data sections contiguous or non-contiguous to identify spatial domains across sections.\n- Scalable to different data resolutions, i.e., wild range of technologies and platforms of SRT data, including **Visium HD**, **Visum**, **MERFISH**, **STARmap**, **Stereo-seq**, **ST**, etc.\n- Scalable to large datasets with a high number of cells and spatial locations.\n- Performance of integrative analysis across modalities (scRNA-seq and SRT) and cell-type deconvolution for the non-single-cell resolution SRT data.\n\n## Other Capabilities\n- Capability to produce not only the batch-corrected embeddings but also batch-corrected gene expression profiles for scRNA-seq data.\n- Capability to perform spatial mapping of reference scRNA-seq data points to the spatial locations of SRT data based on the learned co-embeddings and kNN.\n- Comprehensive `adata` processing by specifically desinged `BaseDataset` class and its view version `MaskedDataset` class for the SRT data, which can be easily integrated with the PyTorch `DataLoader` class for training and validation.\n- Low computational cost and high efficiency in processing large-scale SRT data: 4-8 GB GPU memory is sufficient for processing a dataset with 100,000 spatial locations with 2000+ sample-size in fast mode, and consumes less than 6 minutes for training in 2000 iterations (tested on NVIDIA RTX 3090).\n\n## System Requiremtns\n- **Software Requirements**: Python 3.10+, CUDA 11.6+, PyTorch 1.13.1+, and dgl 1.1.3+.\n- **Hardware Requirements**: NVIDIA GPU with CUDA support (recommended), 8GB+ GPU memory. As possible as high RAM and CPU cores for storing and processing large-scale data (this is not required by STEP, but by the data itself).\n## Installation\n\n### Quick Installation (Recommended for Python 3.10, Linux, CUDA 11.7)\n\n```bash\npip install step-kit[cu117]\n```\n\nThis command installs STEP along with compatible versions of PyTorch and DGL.\n\n### Basic Installation\n\n```bash\npip install step-kit\n```\n\n\u26a0\ufe0f **Critical Note**: \n- This basic installation **does not** include PyTorch and DGL.\n- STEP **will not function** without PyTorch and DGL properly installed.\n- You must install these dependencies separately before using STEP.\n\n### Manual Dependency Installation\n\nIf you need to install specific versions of PyTorch and DGL:\n\n1. Install DGL (example for latest DGL compatible with PyTorch 2.3.x and CUDA 11.8):\n ```bash\n pip install dgl -f https://data.dgl.ai/wheels/torch-2.3/cu118/repo.html\n ```\n\n2. Install PyTorch (example for PyTorch 2.3.1 with CUDA 11.8):\n ```bash\n pip install torch==2.3.1+cu118 -f https://download.pytorch.org/whl/cu118/torch_stable.html\n ```\n\nFor more information on installing PyTorch and DGL, visit:\n- PyTorch: https://pytorch.org/\n- DGL: https://www.dgl.ai/\n\nRemember to choose versions compatible with your system configuration and CUDA version (if applicable).\n\n## Contribution\n\nWe welcome contributions! Please see [`CONTRIBUTING.md`](./CONTRIBUTING.md) for more details!\n\n## License\n\nstep is licensed under [LICENSE](./LICENSE)\n\n## Contact\n\nIf you have any questions, please feel free to contact us at [here](mailto:lilounan1997@gmail.com), or feel free to open an issue on this repository.\n\n## Citation\nThe preprint of STEP is available at [bioRxiv](https://www.biorxiv.org/content/early/2024/04/20/2024.04.15.589470.full.pdf). If you use STEP in your research, please cite:\n\n```bibtex\n@article{Li2024.04.15.589470,\n title = {{{STEP}}: {{Spatial}} Transcriptomics Embedding Procedure for Multi-Scale Biological Heterogeneities Revelation in Multiple Samples},\n author = {Li, Lounan and Li, Zhong and Li, Yuanyuan and Yin, Xiao-ming and Xu, Xiaojiang},\n year = {2024},\n journal = {bioRxiv : the preprint server for biology},\n eprint = {https://www.biorxiv.org/content/early/2024/04/20/2024.04.15.589470.full.pdf},\n publisher = {Cold Spring Harbor Laboratory},\n doi = {10.1101/2024.04.15.589470},\n}\n```\n",
"bugtrack_url": null,
"license": "Apache-2.0",
"summary": "STEP, an acronym for Spatial Transcriptomics Embedding Procedure, is a deep learning-based tool for the analysis of single-cell RNA (scRNA-seq) and spatially resolved transcriptomics (SRT) data. STEP introduces a unified approach to process and analyze multiple samples of scRNA-seq data as well as align several sections of SRT data, disregarding location relationships. Furthermore, STEP conducts integrative analysis across different modalities like scRNA-seq and SRT.",
"version": "0.2.6",
"project_urls": {
"Documentation": "https://sggb0nd.github.io/step/",
"Homepage": "https://github.com/SGGb0nd/step",
"Repository": "https://github.com/SGGb0nd/step"
},
"split_keywords": [
"spatial transcriptomics",
" single-cell rna-seq",
" deep learning",
" scrna-seq",
" srt",
" step",
" step"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "4e28580ab7780f1b8711a9df9fd7dbcac794080d3051a941ee6f1e16dff40ea9",
"md5": "cba7476a37b1b1873bb52b1dcb73f3f4",
"sha256": "00659bd429c04f06ec692316b1c179c939ffa38064e65e8d8fb0580a22899eae"
},
"downloads": -1,
"filename": "step_kit-0.2.6-py3-none-any.whl",
"has_sig": false,
"md5_digest": "cba7476a37b1b1873bb52b1dcb73f3f4",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": "<4.0,>=3.10",
"size": 99165,
"upload_time": "2024-11-05T05:13:47",
"upload_time_iso_8601": "2024-11-05T05:13:47.422828Z",
"url": "https://files.pythonhosted.org/packages/4e/28/580ab7780f1b8711a9df9fd7dbcac794080d3051a941ee6f1e16dff40ea9/step_kit-0.2.6-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "a41a59da27dcb04290f526a2f5e07ce9d5122515456f369268c46c01b292da11",
"md5": "301a398622eceab152cc2f716dcbb23a",
"sha256": "5ff41ac72369c978f2b2e0e5a0046f2abc1391db534b50ee1667fcbb18842b84"
},
"downloads": -1,
"filename": "step_kit-0.2.6.tar.gz",
"has_sig": false,
"md5_digest": "301a398622eceab152cc2f716dcbb23a",
"packagetype": "sdist",
"python_version": "source",
"requires_python": "<4.0,>=3.10",
"size": 85343,
"upload_time": "2024-11-05T05:13:49",
"upload_time_iso_8601": "2024-11-05T05:13:49.289644Z",
"url": "https://files.pythonhosted.org/packages/a4/1a/59da27dcb04290f526a2f5e07ce9d5122515456f369268c46c01b292da11/step_kit-0.2.6.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-11-05 05:13:49",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "SGGb0nd",
"github_project": "step",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"lcname": "step-kit"
}
