Kernelized multiview signed graph learning for single-cell RNA sequencing data.

Area of Science:

• Systems Biology
• Computational Biology
• Genomics

Background:

• Gene regulatory network (GRN) topology is crucial in systems biology.
• Single-cell technologies enable finer GRN resolution but present challenges due to data sparsity and heterogeneity.
• Existing GRN methods often fail with single-cell data and cannot model multiple conditions simultaneously.

Purpose of the Study:

• To develop a novel method for reconstructing GRNs from single-cell data across multiple conditions.
• To address limitations of existing methods in handling cellular heterogeneity and multiple datasets.

Main Methods:

• Proposed single-cell Multi-network Signed Graph Learning (scMSGL), a graph signal processing-based approach.
• Modeled GRNs and gene expressions as signed graphs and graph signals, respectively.
• Jointly estimated multiple GRNs by optimizing gene expression total variation and enforcing similarity through a consensus graph.

Main Results:

• scMSGL demonstrated superior performance in GRN recovery on simulated datasets compared to state-of-the-art methods.
• Successfully identified known gene regulators in mouse embryonic stem cell differentiation.
• Applied successfully to a medulloblastoma cancer clinical study.

Conclusions:

• scMSGL is an effective tool for reconstructing gene regulatory networks from single-cell data, especially across multiple conditions.
• The method advances the analysis of cellular heterogeneity in systems biology.
• scMSGL has practical applications in understanding developmental processes and cancer biology.