Bi-EB: Empirical Bayesian Biclustering for Multi-Omics Data Integration Pattern Identification among Species.

Area of Science:

• Bioinformatics and Computational Biology
• Genomics and Systems Biology
• Translational Cancer Research

Background:

• Current biclustering algorithms have limited application in cross-species, multi-omics data analysis for pattern identification.
• Translational research faces challenges in linking genotype variations to phenotypes and applying findings to patient subpopulations.

Purpose of the Study:

• To develop a novel biclustering algorithm (Bi-EB) for detecting shared patterns in integrated omics data across species.
• To address the clinical need for identifying genotype-phenotype modules and translating findings from cancer cell screening to patient data.

Main Methods:

• Developed a fast empirical Bayesian biclustering (Bi-EB) algorithm incorporating probabilistic interpretation and ratio strategy.
• Utilized an expectation-maximization (EM) algorithm with adjustable parameters (bicluster membership probability threshold Ac, average probability p).
• Validated Bi-EB using simulations and real mRNA/protein data from The Cancer Genomics Atlas (TCGA) and The Cancer Cell Line Encyclopedia (CCLE).

Main Results:

• Bi-EB significantly improved clustering recovery (0.98) and relevance accuracy (0.99) compared to seven other methods.
• Identified shared mRNA-protein causality patterns in breast cancer subtypes (luminal-like and basal-like) from TCGA and CCLE data.
• Discovered key therapeutic target modules (e.g., ER, AR) and novel gene sets (e.g., CCNB1, CDH1) for specific subtypes.

Conclusions:

• The Bi-EB algorithm is a powerful tool for discovering cross-patterns in multi-omics and cross-species data.
• Bi-EB facilitates translational research by bridging cancer cell screening findings with clinical patient data.
• Clinical implementation of Bi-EB can guide targeted therapies and improve patient stratification.