A supervised weighted similarity measure for gene expressions using biological knowledge

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Analyzing gene expression data requires robust similarity measures.
• Existing methods like Pearson correlation and Euclidean distance have limitations with multiple experimental conditions.
• Accurate gene similarity is crucial for understanding biological pathways and functions.

Purpose of the Study:

• To develop a supervised similarity measure for Saccharomyces cerevisiae gene expression data.
• To improve gene function prediction by integrating diverse experimental conditions.
• To introduce Weighted Pearson Correlation (WPC) as a superior similarity metric.

Main Methods:

• Developed Weighted Pearson Correlation (WPC) by optimizing weights using GO-Slim process annotations from SGD.
• Determined weights by maximizing positive predictive value (PPV) for gene pairs with Pearson correlation > 0.80.
• Clustered genes using k-medoid with WPC and predicted functions for unclassified genes using MIPS annotations.

Main Results:

• Successfully predicted functions for 55 Saccharomyces cerevisiae genes with high confidence (p-value > 10^-10).
• Demonstrated the superiority of WPC over Pearson correlation and Euclidean distance using PPV.
• Identified and functionally categorized 135 unclassified genes.

Conclusions:

• Weighted Pearson Correlation (WPC) is an effective supervised similarity measure for gene expression data.
• WPC enhances gene function prediction accuracy, especially with multiple experimental conditions.
• The developed method provides a valuable tool for yeast genomics research.