SLIVER: Unveiling large scale gene regulatory networks of single-cell transcriptomic data through causal structure...

Area of Science:

• Computational Biology and Bioinformatics
• Systems Biology
• Genomics and Transcriptomics

Background:

• Current gene regulatory network (GRN) inference methods predominantly use correlation analysis, which fails to capture the inherent causal relationships in biological systems.
• Existing causal discovery algorithms, often based on Directed Acyclic Graphs (DAGs), face computational challenges with large networks, limiting their scalability.

Purpose of the Study:

• To develop a scalable and accurate causal inference algorithm for gene regulatory network construction.
• To address the limitations of existing methods in handling complex biological data and large-scale networks.

Main Methods:

• Proposed the SLIVER (causal Discovery Via dimensionality Reduction) algorithm, integrating causal structural equation models and graph decomposition.
• SLIVER utilizes factor nodes to represent functional modules, reducing the GRN to low-dimensional matrices for efficient causal learning.
• Employed structural causal models (SCM) and enforced DAG constraints in the reduced dimensional space, guiding functional aggregation via cosine similarity.

Main Results:

• SLIVER demonstrated superior performance in GRN inference accuracy and computational efficiency across 12 single-cell transcriptomic datasets compared to 12 widely used methods.
• Analysis of factor nodes revealed biological explanations for gene aggregation within functional modules.
• Successful application to Type 2 diabetes mellitus scRNA-seq data identified key transcriptional regulatory changes in β cells.

Conclusions:

• SLIVER offers a robust and scalable approach for causal GRN inference, overcoming limitations of traditional methods.
• The algorithm provides biologically interpretable insights into gene regulation and functional modules.
• SLIVER holds promise for advancing our understanding of complex diseases like Type 2 diabetes mellitus at the molecular level.