High-Order Correlation Integration for Single-Cell or Bulk RNA-seq Data Analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate sample type quantification and labeling are crucial for understanding complex diseases.
• Data noise reduction and preservation of intrinsic patterns are vital for reliable sample clustering and classification.
• Existing methods struggle with high-dimensional and heterogeneous omics data.

Purpose of the Study:

• To introduce High-order Correlation Integration (HCI), a novel data integration framework for high-dimensional data feature extraction.
• To improve the accuracy and robustness of sample clustering and classification algorithms.
• To demonstrate HCI's effectiveness in analyzing diverse biological datasets.

Main Methods:

• HCI utilizes high-order correlation matrices and pattern fusion analysis (PFA) for feature extraction.
• High-order Pearson's correlation coefficients identify latent patterns in noisy datasets.
• PFA efficiently extracts intrinsic sample patterns from multiple data matrices.

Main Results:

• HCI accurately identified cell types from single-cell RNA-seq data with superior accuracy and robustness compared to existing methods.
• HCI outperformed other methods in identifying distinct cancer subtypes when integrating heterogeneous omics data from TCGA and GEO.
• HCI-derived mRNA-miRNA regulatory networks for colorectal cancer revealed significant enrichment in known functional pathways and disease annotations.

Conclusions:

• HCI is an effective framework for high-dimensional data integration and feature extraction.
• HCI demonstrates broad applicability and flexibility for sample clustering across various RNA-seq data types and organizations.
• The framework shows promise for advancing disease subtype identification and understanding molecular mechanisms.