delnx 0.2.1

Differential expression analysis for single-cell genomics

🌳 delnx

delnx ("de-lo-nix" | /dɪˈlɒnɪks/) is a python package for differential expression analysis of (single-cell) genomics data. It enables scalable analyses of atlas-level datasets through GPU-accelerated regression models and statistical tests implemented in JAX. It also provides a consistent interface to perform DE analysis with other methods, such as statsmodels and PyDESeq2.

🚀 Installation

PyPI

pip install delnx

Development version

pip install git+https://github.com/joschif/delnx.git@main

⚡ Quickstart

import delnx as dx

# Compute size factors
adata = dx.pp.size_factors(adata, method="ratio")

# Estimate dispersion parameters
adata = dx.pp.dispersion(adata, size_factor_key="size_factor", method="deseq2")

# Run differential expression analysis
results = dx.tl.de(
    adata,
    condition_key="condition",
    group_key="cell_type",
    mode="all_vs_ref",
    reference="control",
    method="negbinom",
    size_factor_key="size_factor",
    dispersion_key="dispersion",
)

💎 Features

• Pseudobulking: Perform DE on large multi-sample datasets by using pseudobulk aggregation.
• Size factor estimation: Compute size factors for normalization and DE analysis.
• Dispersion estimation: Estimate dispersion parameters for negative binomial models.
• Differential expression analysis: Consistent interface to perform DE analysis using various methods, including:
  • Negative binomial regression with dispersion estimates.
  • Logistic regression with a likelihood ratio test.
  • ANOVA tests based on linear models.
  • DESeq2 through PyDESeq2, a widely used method for DE analysis of RNA-seq data.
• GPU acceleration: Most methods are implemented in JAX, enabling GPU acceleration for scalable DE-analysis on large datasets.

⚙️ Backends

delnx implements DE tests using regression models and statistical tests from various backends:

• JAX
• statsmodels
• cuML
• PyDESeq2

🗺️ Roadmap

• Provide a common interface to standard GLM-based DE tests (inspired by Seurat::FindMarkers)
  • Logistic regression and likelihood ratio test
    • statsmodels
    • JAX
    • cuML
  • Negative binomial regression
    • statsmodels
    • JAX
  • ANOVA
    • statsmodels
    • JAX
  • Binomial regression for binary data
    • statsmodels
    • JAX
• Implement DESeq2 wrapper using PyDESeq2
• Implement size factor estimation methods
• Add dispersion estimation methods
  • Basic gene-wise dispersion estimation
  • DESeq2 and edgeR-inspired dispersion estimation with shrinkage
  • Take design and covariates into account for dispersion estimation
• Plotting functions to visualize DE results
• Gene set enrichment analysis for DE results

📖 Documentation

For more information, check out the documentation and the API reference.