Hurdle

Hurdle models zero-inflated multivariate single-cell gene expression data to enable inference of gene co-regulatory networks.

Key Features:

• Multivariate Hurdle Model: Implements a multivariate Hurdle model that modifies standard distributions to account for zero-inflation in single-cell gene expression data.
• Singular Gaussian Distributions: Uses a mixture of singular Gaussian distributions to model the conditional Normal distribution with singularities along coordinate axes, generalizing univariate zero-inflated models to higher dimensions.
• Neighborhood Selection and Graphical Models: Employs neighborhood selection using pseudo-likelihood with a group lasso penalty to select and fit undirected graphical models that capture conditional independences between genes.
• Sensitivity and Robustness: Shows increased sensitivity and robustness to deviations from the Hurdle model structure, improving detection of network structures in complex datasets.

Scientific Applications:

• Gene Co-regulatory Network Inference: Enables inference of gene co-regulatory networks from zero-inflated single-cell expression data to elucidate regulatory mechanisms within cell types.
• Single-Cell Data Analysis: Applies to single-cell datasets, including T follicular helper cells and mouse dendritic cells, where it has identified network structures not previously detected.

Methodology:

Implements parameter estimation and sampling routines for multivariate Hurdle models based on a Normal density; models conditional distributions via a mixture of singular Gaussian distributions with coordinate-axis singularities; and applies neighborhood selection using pseudo-likelihood with a group lasso penalty to fit undirected graphical models.