Constructing gene co-expression networks and predicting functions of unknown genes by random matrix theory

Area of Science:

• Genomics and Systems Biology
• Bioinformatics and Computational Biology

Background:

• Genome-scale sequencing enables the study of complex cellular networks.
• Current methods for defining gene co-expression networks from microarray data lack robustness and consistency.
• Identifying true correlations amidst noise is challenging, especially for less-studied organisms.

Purpose of the Study:

• To apply Random Matrix Theory (RMT) to microarray data for robustly defining gene co-expression networks.
• To establish a theoretical criterion for determining correlation thresholds in biological systems.
• To leverage RMT for gene function prediction in understudied organisms.

Main Methods:

• Characterization of correlation matrices from microarray profiles using RMT.
• Analysis of the nearest neighbor spacing distribution (NNSD) of correlation matrices.
• Application of RMT to diverse species including yeast, E. coli, and human.

Main Results:

• A sharp transition in NNSD was identified, enabling the determination of a reliable correlation threshold.
• Modular co-expression networks were successfully revealed in yeast cell cycle data, aligning with gene annotations.
• The RMT approach demonstrated sensitivity, robustness, and reliability in gene function prediction for unknown genes.

Conclusions:

• The transition of NNSD in correlation matrices offers a profound theoretical criterion for identifying gene co-expression networks.
• RMT provides a robust framework for dissecting cellular networks and predicting gene functions.
• This method enhances the analysis of genome-wide expression data across various biological systems.