PLNFGL: joint estimation of multi-condition gene networks from single-cell RNA-seq data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gaussian graphical models (GGMs) are limited for single-cell RNA sequencing (scRNA-seq) due to data characteristics like dropout events and distributional mismatch.
• Existing network inference methods often analyze single conditions, restricting their application in multi-condition biological studies.
• There is a need for robust methods to infer gene interaction networks from scRNA-seq data, especially across different conditions.

Purpose of the Study:

• To propose PLNFGL (Poisson Log-Normal Fused Graphical Lasso), a novel framework for joint network estimation in scRNA-seq data.
• To develop a method that accommodates dropout effects and distributional properties of scRNA-seq data.
• To enable the inference of condition-specific gene interaction networks for multi-condition studies.

Main Methods:

• Utilized a multivariate Poisson log-normal model to handle dropout effects and distributional characteristics of scRNA-seq data.
• Employed moment methods for covariance estimation.
• Applied a joint graphical model to infer condition-specific precision matrices, enabling network comparison across conditions.

Main Results:

• PLNFGL demonstrated improved estimation accuracy in simulation studies compared to existing methods.
• Applied to Alzheimer's disease and lung cancer spatial transcriptomics data, PLNFGL revealed cell-type-specific interaction networks.
• Edge set enrichment analysis identified known interactions and highlighted novel disease-related gene targets.

Conclusions:

• PLNFGL provides a powerful and accurate tool for the integrative analysis of scRNA-seq data.
• The method facilitates the discovery of condition-specific gene interaction networks, advancing biological insights.
• This framework supports pathway analysis and the identification of novel therapeutic targets in complex diseases.