Echidna

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.

Install

Echidna is available on PyPI under the name sc-echidna and bioconda 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.

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 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.
lkj_concentration	float	1.0	Concentration parameter of LKJ prior. Values > 1.0 result in more diagonal covariance matrices.
q_shape_rate_scaler	float	10.0	Scaler for the shape and rate parameters of the covariance diagonal for variational inference.
q_corr_init	float	0.01	Initial scale of the variational correlation.
q_cov_scaler	float	0.01	Scaler for the covariance of the variational correlation.
eta_mean_init	float	2.0	Initial mean value for the eta parameter.
eps	float	5e-3	Small constant added to the diagonal to ensure positive definiteness (PD).