A causal reinforcement learning framework for reliable gene regulatory network inference.

Area of Science:

• Computational Biology
• Systems Biology
• Genomics

Background:

• Gene regulatory network inference is vital for understanding cellular functions and disease mechanisms.
• Traditional methods lack causal directionality and interpretability, hindering accurate pattern capture in large networks.
• Existing deep learning approaches struggle with structural interpretability and efficient large-scale network optimization.

Purpose of the Study:

• To develop a gene regulatory network inference method addressing limitations of current approaches.
• To improve causal direction identification and structural interpretability in gene networks.
• To enable efficient global optimization for large-scale gene regulatory network structures.

Main Methods:

• Proposes a novel gene regulatory network inference method utilizing causal reinforcement learning.
• Applies the method to benchmark datasets (DREAM5) and single-cell RNA sequencing (scRNA-seq) data.
• Employs GNNExplainer for interpretability analysis and KEGG pathway enrichment for biological validation.

Main Results:

• The proposed method surpasses baseline models in inference accuracy, structural sparsity, and biological plausibility.
• The inferred gene regulatory networks demonstrate clarified causal relationships and enhanced structural interpretability.
• Topological and functional analyses show strong alignment with real biological networks and known pathways.

Conclusions:

• The causal reinforcement learning framework enables efficient global optimization of large-scale gene networks.
• Inferred networks exhibit high causal rationality and biological interpretability, validating the method's effectiveness.
• Presents a scalable approach for causal structure modeling of high-dimensional gene expression data in systems and computational biology.