# Echidna
<img src=./echidna_logo.png width="100" />
A Bayesian framework for quantifying gene dosage effect on phenotypic plasticity through integrating single-cell RNA sequencing (scRNA-seq) and bulk whole-genome sequencing (WGS) from a single or multiple time points.
<img src=./echidna_concept.png width="800" />
# Install
Echidna is available on [PyPI](https://pypi.org/project/sc-echidna/1.0.0/) under the name `sc-echidna` and [bioconda](https://anaconda.org/bioconda/echidna) under `echidna`.
## Step 1 (optional but recommended)
Create a conda environment with a recent Python version: `conda create -n "echidna-env" python=3.10`.
## Step 2
Ensure you have the right version of torch installed for your device. See instructions on [pytorch.org](https://pytorch.org/get-started/locally/).
### Conda Formula
`conda install bioconda::echidna`
### Pip Formula
`pip install sc-echidna`
Installation time depends on hardware, but can be finished within 5 minutes for most computers.
# Tutorial
There are four example notebooks:
1. 1-single-timepoint.ipynb
2. 2-multi-timepoint.ipynb
3. 3-infer-gene-dosage.ipynb
4. 4-echidna-model.ipynb
The notebooks are meant to be run sequentially, and they build off of each other.
Notebook 1 introduces you to the package - preparing your data, setting hyperparamters, performing posterior predictive checks - with data collected from a single point in time.
In notebook 2, we look at a multi-timepoint setting, where we have paired single-cell and WGS data collected over time. The demo dataset included in ./demo_data is the same data we used for this notebook.
The saved model runs from notebook 2 will be used in notebook 3, where you will see how to infer amplifications and deletions by cluster of genes across a given genome. This notebook also shows you how to calculate and plot gene dosage effect with Echidna.
Notebook 4 is meant to show you how to do more custom work with the model. We package together many functions for your convenience, but this notebook will show you how to work directly with the model for the experiments not covered in the package. Some [Pyro](https://pyro.ai/) knowledge is assumed.
# Echidna Configuration Settings
## `.obs` Labels
| Setting | Type | Default | Description |
|----------------|--------|-----------------|-------------------------------------------|
| `timepoint_label` | `str` | `"timepoint"` | Label for timepoints in the data. |
| `counts_layer` | `str` | `"counts"` | Name of the counts layer in the data. |
| `clusters` | `str` | `"leiden"` | Clustering method used in the data. This can also be celltype annotations, if you have them. |
## Training Parameters
| Setting | Type | Default | Description |
|-----------------|--------|-----------------|-----------------------------------------------------------|
| `seed` | `int` | `42` | Random seed for reproducibility. |
| `n_steps` | `int` | `10000` | Maximum number of steps for Stochastic Variational Inference (SVI). |
| `learning_rate` | `float`| `0.1` | Learning rate for the Adam optimizer. |
| `val_split` | `float`| `0.1` | Percentage of training data to use for validation. |
| `patience` | `int` | `30` | Early stopping patience (set to >0 to enable early stopping). |
| `device` | `str` | `"cuda" if is_available() else "cpu"` | Device to use for training (GPU if available, otherwise CPU). |
| `verbose` | `bool` | `True` | Whether to enable logging output. |
## Model Hyperparameters
| Setting | Type | Default | Description |
|----------------------|---------|-----------|------------------------------------------------------------------------------------|
| `inverse_gamma` | `bool` | `False` | Whether to use inverse gamma for noisier data. |
| `eta_mean_init` | `float` | `2.0` | Initial mean value for the eta parameter. |
| `lkj_concentration` | `float` | `1.0` | Concentration parameter of LKJ prior. Values > 1.0 result in more diagonal covariance matrices. |
| `q_corr_init` | `float` | `0.01` | Initial scale of the variational correlation. |
| `q_shape_rate_scaler` | `float` | `10.0` | Scaler for the shape and rate parameters of the covariance diagonal for variational inference. |
Raw data
{
"_id": null,
"home_page": null,
"name": "sc-echidna",
"maintainer": null,
"docs_url": null,
"requires_python": "<4.0,>=3.10",
"maintainer_email": null,
"keywords": "scRNA, wgs, cnv",
"author": "Joy Linyue Fan",
"author_email": "lf2684@columbia.edu",
"download_url": "https://files.pythonhosted.org/packages/3b/77/c2036819dd48bc222ba673185041999cdccbe3957c217d988c94fab85966/sc_echidna-1.0.3.tar.gz",
"platform": null,
"description": "# Echidna\n<img src=./echidna_logo.png width=\"100\" />\n\n A Bayesian framework for quantifying gene dosage effect on phenotypic plasticity through integrating single-cell RNA sequencing (scRNA-seq) and bulk whole-genome sequencing (WGS) from a single or multiple time points. \n\n\n<img src=./echidna_concept.png width=\"800\" />\n\n# Install\n\nEchidna is available on [PyPI](https://pypi.org/project/sc-echidna/1.0.0/) under the name `sc-echidna` and [bioconda](https://anaconda.org/bioconda/echidna) under `echidna`.\n\n## Step 1 (optional but recommended)\nCreate a conda environment with a recent Python version: `conda create -n \"echidna-env\" python=3.10`.\n\n## Step 2\nEnsure you have the right version of torch installed for your device. See instructions on [pytorch.org](https://pytorch.org/get-started/locally/).\n\n### Conda Formula\n\n`conda install bioconda::echidna`\n\n### Pip Formula\n`pip install sc-echidna`\n\nInstallation time depends on hardware, but can be finished within 5 minutes for most computers.\n\n# Tutorial\n\nThere are four example notebooks:\n\n1. 1-single-timepoint.ipynb\n2. 2-multi-timepoint.ipynb\n3. 3-infer-gene-dosage.ipynb\n4. 4-echidna-model.ipynb\n\nThe notebooks are meant to be run sequentially, and they build off of each other.\n\nNotebook 1 introduces you to the package - preparing your data, setting hyperparamters, performing posterior predictive checks - with data collected from a single point in time.\n\nIn notebook 2, we look at a multi-timepoint setting, where we have paired single-cell and WGS data collected over time. The demo dataset included in ./demo_data is the same data we used for this notebook.\n\nThe saved model runs from notebook 2 will be used in notebook 3, where you will see how to infer amplifications and deletions by cluster of genes across a given genome. This notebook also shows you how to calculate and plot gene dosage effect with Echidna.\n\nNotebook 4 is meant to show you how to do more custom work with the model. We package together many functions for your convenience, but this notebook will show you how to work directly with the model for the experiments not covered in the package. Some [Pyro](https://pyro.ai/) knowledge is assumed.\n\n# Echidna Configuration Settings\n\n## `.obs` Labels\n\n| Setting | Type | Default | Description |\n|----------------|--------|-----------------|-------------------------------------------|\n| `timepoint_label` | `str` | `\"timepoint\"` | Label for timepoints in the data. |\n| `counts_layer` | `str` | `\"counts\"` | Name of the counts layer in the data. |\n| `clusters` | `str` | `\"leiden\"` | Clustering method used in the data. This can also be celltype annotations, if you have them. |\n\n## Training Parameters\n\n| Setting | Type | Default | Description |\n|-----------------|--------|-----------------|-----------------------------------------------------------|\n| `seed` | `int` | `42` | Random seed for reproducibility. |\n| `n_steps` | `int` | `10000` | Maximum number of steps for Stochastic Variational Inference (SVI). |\n| `learning_rate` | `float`| `0.1` | Learning rate for the Adam optimizer. |\n| `val_split` | `float`| `0.1` | Percentage of training data to use for validation. |\n| `patience` | `int` | `30` | Early stopping patience (set to >0 to enable early stopping). |\n| `device` | `str` | `\"cuda\" if is_available() else \"cpu\"` | Device to use for training (GPU if available, otherwise CPU). |\n| `verbose` | `bool` | `True` | Whether to enable logging output. |\n\n## Model Hyperparameters\n\n| Setting | Type | Default | Description |\n|----------------------|---------|-----------|------------------------------------------------------------------------------------|\n| `inverse_gamma` | `bool` | `False` | Whether to use inverse gamma for noisier data. |\n| `eta_mean_init` | `float` | `2.0` | Initial mean value for the eta parameter. |\n| `lkj_concentration` | `float` | `1.0` | Concentration parameter of LKJ prior. Values > 1.0 result in more diagonal covariance matrices. |\n| `q_corr_init` | `float` | `0.01` | Initial scale of the variational correlation. |\n| `q_shape_rate_scaler` | `float` | `10.0` | Scaler for the shape and rate parameters of the covariance diagonal for variational inference. |\n\n",
"bugtrack_url": null,
"license": "MIT",
"summary": "Mapping genotype to phenotype through joint probabilistic modeling of single-cell gene expression and chromosomal copy number variation.",
"version": "1.0.3",
"project_urls": {
"Homepage": "https://github.com/azizilab/echidna",
"Repository": "https://github.com/azizilab/echidna"
},
"split_keywords": [
"scrna",
" wgs",
" cnv"
],
"urls": [
{
"comment_text": null,
"digests": {
"blake2b_256": "2315e571585b42475ee64a439169987f62935df1cd4a18c4077053e6e5961244",
"md5": "22ace14d4f878742e0564fda87fc942b",
"sha256": "63291b06a534802bf7c1ee580d7f5c0342e3f8f321b7ee5bda7529dcd17d1f3c"
},
"downloads": -1,
"filename": "sc_echidna-1.0.3-py3-none-any.whl",
"has_sig": false,
"md5_digest": "22ace14d4f878742e0564fda87fc942b",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": "<4.0,>=3.10",
"size": 40523,
"upload_time": "2025-07-24T18:39:44",
"upload_time_iso_8601": "2025-07-24T18:39:44.463666Z",
"url": "https://files.pythonhosted.org/packages/23/15/e571585b42475ee64a439169987f62935df1cd4a18c4077053e6e5961244/sc_echidna-1.0.3-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": null,
"digests": {
"blake2b_256": "3b77c2036819dd48bc222ba673185041999cdccbe3957c217d988c94fab85966",
"md5": "2af575398291948bfb9217b40af50349",
"sha256": "7ba8c0bbe4c39f09f0dad36413a7b9d06db0dc10d270e43392634cc822400aaa"
},
"downloads": -1,
"filename": "sc_echidna-1.0.3.tar.gz",
"has_sig": false,
"md5_digest": "2af575398291948bfb9217b40af50349",
"packagetype": "sdist",
"python_version": "source",
"requires_python": "<4.0,>=3.10",
"size": 35028,
"upload_time": "2025-07-24T18:39:45",
"upload_time_iso_8601": "2025-07-24T18:39:45.690075Z",
"url": "https://files.pythonhosted.org/packages/3b/77/c2036819dd48bc222ba673185041999cdccbe3957c217d988c94fab85966/sc_echidna-1.0.3.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-07-24 18:39:45",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "azizilab",
"github_project": "echidna",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"requirements": [
{
"name": "scanpy",
"specs": [
[
">=",
"1.10"
]
]
},
{
"name": "pandas",
"specs": []
},
{
"name": "numpy",
"specs": [
[
">=",
"1.21"
]
]
},
{
"name": "seaborn",
"specs": []
},
{
"name": "hmmlearn",
"specs": []
},
{
"name": "pyro-ppl",
"specs": [
[
">=",
"1.9.1"
]
]
},
{
"name": "graphviz",
"specs": []
},
{
"name": "requests",
"specs": []
},
{
"name": "scikit-learn",
"specs": []
},
{
"name": "leidenalg",
"specs": []
}
],
"lcname": "sc-echidna"
}
JSON
