Gene ranking and biomarker discovery under correlation

Area of Science:

• Genomics
• Biostatistics
• Bioinformatics

Background:

• Standard gene ranking methods in high-throughput genomics often overlook gene-gene correlations.
• Ignoring these correlations can significantly impact gene ordering and the power of statistical tests.

Purpose of the Study:

• To introduce a novel method for gene ranking that incorporates gene-gene correlations.
• To improve the accuracy of biomarker discovery and gene set evaluation.

Main Methods:

• Developed correlation-adjusted t-scores (cat scores) by adjusting gene-wise t-statistics for correlations.
• Proposed a shrinkage procedure for computing cat scores from small sample data.
• Utilized a predictive approach for variable selection in two-class linear discriminant analysis.

Main Results:

• The cat score method was shown to improve gene ordering estimation compared to standard t-scores.
• Demonstrated increased statistical power for a fixed true discovery rate.
• Successfully applied the cat score to analyze metabolomic data.

Conclusions:

• The correlation-adjusted t-score (cat score) offers a more robust approach to gene ranking in genomic studies.
• The method is particularly beneficial when gene-gene correlations are present.
• The cat score is implemented in the freely available R package 'st'.