Nonlinear Network Reconstruction from Gene Expression Data Using Marginal Dependencies Measured by DCOL.

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Network reconstruction from high-throughput expression data is crucial for identifying regulatory relationships.
• Biological systems exhibit complex, nonlinear interactions often missed by linear association methods.
• The Distance based on Conditional Ordered List (DCOL) is a sensitive and efficient measure for nonlinear relations.

Purpose of the Study:

• To evaluate the utility of DCOL in network reconstruction by integrating it with local false discovery rate (lfdr)-based inference.
• To introduce a new method, nlnet, for uncovering nonlinear regulatory modules.
• To demonstrate the effectiveness of nlnet in both simulated and real-world biological data.

Main Methods:

• Combining the Distance based on Conditional Ordered List (DCOL) with local false discovery rate (lfdr)-based inference.
• Network reconstruction using high-throughput expression data.
• Validation through simulations and single-cell RNA sequencing (scRNA-seq) data analysis.

Main Results:

• The nlnet method effectively recovers hidden nonlinear modules in simulated data.
• nlnet demonstrates utility in analyzing a single-cell RNA sequencing dataset.
• The approach leverages nonlinear dependencies for more comprehensive network insights.

Conclusions:

• nlnet provides a powerful tool for biological network reconstruction, particularly for uncovering complex nonlinear interactions.
• The method enhances the identification of regulatory relationships missed by linear methods.
• nlnet is available as an R package, facilitating its application in biological research.