Scalable cell-specific coexpression networks for granular regulatory pattern discovery with NeighbourNet

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Gene networks are crucial for understanding gene expression regulation.
• Single-cell RNA sequencing (scRNA-seq) allows network inference at cellular resolution.
• Existing methods often assume static regulatory programs, missing dynamic cellular variations.

Purpose of the Study:

• Introduce NeighbourNet (NNet), a novel method for constructing cell-specific coexpression networks.
• Address limitations of existing methods in capturing dynamic regulatory variations in scRNA-seq data.
• Provide a scalable framework for analyzing large-scale single-cell datasets.

Main Methods:

• NeighbourNet (NNet) embeds gene expression into a low-dimensional space using principal component analysis.
• Local regression within k-nearest neighbours (KNN) quantifies cell-specific coexpression.
• NNet supports scalable downstream analyses, including meta-network aggregation and prior knowledge integration.

Main Results:

• NNet improves computational efficiency and stabilizes coexpression estimates for scRNA-seq data.
• The method effectively mitigates challenges from data noise, sparsity, and small sample sizes in KNN regression.
• Case studies demonstrate NNet's utility in transcription factor activity prediction, hematopoiesis, and tumor microenvironment analysis.

Conclusions:

• NNet offers a novel framework for exploring cellular variation in gene coexpression.
• The R package integrates seamlessly with existing single-cell analysis workflows.
• NNet enables robust inference of cell-specific regulatory programs from scRNA-seq data.