continuousvi 0.1.8

Analyzing single-cell RNA-seq data with continuous covariates extinding scVI.

ContinuousVI

A Python library for analyzing single-cell RNA-seq data with continuous covariates using scVI.

ContinuousVI extends the popular single-cell Variational Inference (scVI) framework to incorporate one or more continuous factors (like pseudotime or aging metrics) while correcting for batch effects. It provides straightforward APIs for:

• Multiple model training (with different random seeds/initializations)
• Generating latent embeddings (e.g., UMAP, clustering)
• Regression against continuous covariates (linear, polynomial, spline)
• Sampling from the generative model for gene expression distributions

🧬 Key Features

1. Continuous Covariate Support: Include a single continuous factor (e.g., pseudotime) alongside batch/cell-type labels.
2. Multiple Model Training: Train N scVI models with identical hyperparameters but varying seeds, enabling robust downstream analyses.
3. Dimensionality Reduction & Clustering: Obtain latent embeddings, run UMAP or Leiden clustering, and easily visualize results.
4. Gene Expression Sampling: Sample expression parameters (px) from the learned generative models for posterior predictive analyses.
5. Regression Tools: Regress expression levels against the continuous covariate using OLS, polynomial, or spline models (including advanced multi-sampling approaches for uncertainty estimation).

📕 Installation

ContinuousVI will be published on PyPI. Once available, you can install it via:

pip install continuousvi

Or install directly from source:

git clone https://github.com/<your-org>/continuousvi.git
cd continuousvi
pip install .

Pip location: pip

🚀 Quick Usage Example

import scanpy as sc
from continuousvi import ContinuousVI

# Load AnnData
adata = sc.read_h5ad("my_data.h5ad")

# Initialize
vi_setup = ContinuousVI(
    adata=adata,
    batch_key="batch",
    label_key="cell_type",
    continuous_key="pseudotime"
)

# Train multiple models
trained_vi = vi_setup.train(n_train=5, n_latent=30)

# Calculate embeddings (UMAP, clustering)
trained_vi.calc_embeddings(resolution=0.5, n_neighbors=10, n_pcs=30)

# Perform a simple linear regression against the continuous covariate
df_regression = trained_vi.regression(mode="ols")
print(df_regression.head())

🛠️ Developer Guide

🔧 Environment Setup with uv

If you use uv (a command-line tool for managing Python environments), you can set up a development environment as follows:

# Clone the repository
git clone https://github.com/<your-org>/continuousvi.git
cd continuousvi

# Create and activate a new uv environment (example name: 'contvi-dev')
uv new env contvi-dev
uv activate contvi-dev

# Install an editable version of ContinuousVI along with dev requirements
pip install -e .[dev]

Note: The [dev] extra (or similar) could include testing and linting dependencies if specified in setup.cfg or pyproject.toml.

📁 Project Structure

• ContinuousVI: Sets up your AnnData object and trains multiple scVI models.
• TrainedContinuousVI: Manages trained models, provides methods for embeddings, regression, and sampling.
• Utility Methods: Perform regression (linear, polynomial, spline), advanced regression with multi-sampling, and more.

🪄 Contributing

1. Fork the repository and create your feature branch from main.
2. Make your changes, ensuring that new code is tested and documented.
3. Create a Pull Request, describing your changes and the reason behind them.

📝 License

ContinuousVI is licensed under the MIT License (or the license relevant to your project). Please see the LICENSE file for details.