Information-incorporated gene network construction with FDR control.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Gene expression studies are crucial for understanding gene networks.
• Partial correlation-based networks are preferred for identifying direct gene associations over marginal correlations.
• Existing methods lack robust False Discovery Rate (FDR) control and integration of prior biological knowledge for partial correlation network construction.

Purpose of the Study:

• To propose a novel method, Partial Correlation Graph with Information Incorporation (PCGII), for constructing gene networks.
• To address limitations in FDR control and the incorporation of prior biological knowledge in partial correlation-based network inference.
• To develop a method capable of handling high-dimensional data.

Main Methods:

• PCGII utilizes regularized node-wise regression to estimate partial correlations between gene pairs.
• The method incorporates prior biological knowledge by controlling for the effects of all other genes.
• It is designed to handle high-dimensional datasets where the number of genes exceeds the sample size and controls FDR.

Main Results:

• PCGII demonstrates superior FDR control and higher statistical power compared to existing methods in simulation studies.
• Application to a plant gene expression dataset successfully identified known regulatory relationships and a central hub gene.
• The method revealed novel direct gene associations, suggesting potential functional relationships within the biological system.

Conclusions:

• PCGII offers an effective approach for constructing gene networks with improved FDR control and the ability to integrate prior biological information.
• The method is suitable for analyzing high-dimensional gene expression data.
• An R package for PCGII is available, along with a pseudogene method for analyses lacking prior information, enabling FDR and power assessment on real data.