Indirect two-sided relative ranking: a robust similarity measure for gene expression data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Vast public gene expression data exists but is underutilized due to experimental variations.
• These variations hinder accurate comparison and integration of diverse gene expression datasets.
• Novel biological insights remain hidden within this fragmented data.

Purpose of the Study:

• To introduce a novel method for comparing gene expression profiles that overcomes experimental variations.
• To enable more comprehensive data integration and facilitate new scientific discoveries.
• To improve the accuracy of downstream analyses using gene expression data.

Main Methods:

• Developed an indirect two-sided relative ranking method for gene expression profile comparison.
• Extended existing correlation-based methods by incorporating rankings across the entire database.
• Validated the method's robustness against experimental barriers like vehicle and batch effects.

Main Results:

• The indirect method significantly improves the retrieval of compounds with similar therapeutic effects across datasets.
• Demonstrated substantial improvements in recall at rank 10 (97.03% and 49.44%) on independent datasets.
• Achieved statistically significant accuracy improvements in classifying cancer subtypes, predicting drug sensitivity, and cell type classification.

Conclusions:

• The indirect two-sided relative ranking method effectively bridges experimental barriers in gene expression data.
• This approach unlocks richer scientific discoveries by enabling the integration of previously incompatible datasets.
• The method offers a significant advancement over existing techniques for gene expression data analysis.