Mutual information for detecting multi-class biomarkers when integrating multiple bulk or single-cell transcriptomic...

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Biomarker detection is crucial in biomedical research.
• Integrating multiple omics studies enhances statistical power but existing methods are limited to two-class scenarios.
• Multi-class omics studies require novel analytical approaches.

Purpose of the Study:

• To propose a statistical framework, Mutual Information Concordance Analysis (MICA), for detecting biomarkers with concordant multi-class expression patterns across multiple omics studies.
• To address limitations of existing methods in handling multi-class omics data.
• To provide a robust method for biomarker discovery in complex biological systems.

Main Methods:

• Developed Mutual Information Concordance Analysis (MICA), an information-theoretic framework.
• Employed a global test using mutual information to detect concordant multi-class patterns.
• Utilized post hoc analysis to identify specific studies exhibiting concordant patterns.

Main Results:

• MICA demonstrated improved accuracy and effective false discovery rate control compared to existing methods in simulations.
• Applied MICA to transcriptomic studies of mouse metabolism and estrogen treatment profiles, revealing biological insights.
• Successfully implemented MICA for single-cell RNA-Seq data, identifying novel tumor progression biomarkers and highlighting the role of ribosomal function in triple-negative breast cancer.

Conclusions:

• MICA is a powerful and accurate statistical framework for biomarker detection in multi-class omics studies.
• The method offers significant advantages over existing approaches for horizontally combining omics data.
• MICA has broad applicability in identifying novel therapeutic targets and understanding complex biological processes.