Mutual information estimation reveals global associations between stimuli and biological processes

Area of Science:

• Systems Biology
• Bioinformatics
• Genomics

Background:

• Microarray gene expression analysis is widely used but interpreting biological changes remains challenging.
• Current methods like gene clustering offer only partial insights into cellular processes.
• A need exists for methods that reveal global cellular changes in response to stimuli or varying conditions.

Purpose of the Study:

• To develop a novel method for discovering global cellular changes by linking gene expression patterns to biological functions.
• To introduce and validate the Least-Squares Mutual Information (LSMI) feature selection method.

Main Methods:

• Developed Least-Squares Mutual Information (LSMI), a novel feature selection technique.
• LSMI computes mutual information without density estimation, enabling detection of nonlinear associations.
• Applied LSMI to analyze yeast microarray datasets.

Main Results:

• LSMI effectively detects nonlinear associations within cellular systems.
• Demonstrated LSMI's superiority over existing methods through comparative analysis.
• Identified that non-natural stimuli impact diverse biological processes, while others show no significant relation.
• Discovered four categories of biological processes based on stimulus response: DNA/RNA metabolism, gene expression, protein metabolism, and protein localization.

Conclusions:

• LSMI is a novel feature selection method for analyzing gene expression data.
• LSMI successfully mined associations between yeast cellular conditions and biological processes using microarray data.
• LSMI provides a powerful tool for elucidating the global organization of cellular process control.