Bayesian generalized biclustering analysis via adaptive structured shrinkage.

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• Biclustering identifies local patterns in data matrices by simultaneously clustering features and samples.
• Existing methods primarily handle continuous data and struggle with diverse omics data types like binomial (SNP) or negative binomial (RNA-seq).
• Current approaches do not leverage biological information, which can enhance variable selection and prediction in omics analyses.

Purpose of the Study:

• To develop a novel Bayesian biclustering method that accommodates multiple data types (Gaussian, Binomial, Negative Binomial).
• To incorporate biological information into the biclustering process for improved feature selection.
• To demonstrate the method's effectiveness and superiority over existing techniques using simulations and multi-omics data.

Main Methods:

• A novel Bayesian biclustering framework was developed.
• The method supports multiple data types, including Gaussian, Binomial, and Negative Binomial distributions.
• A Bayesian adaptive structured shrinkage prior was employed to integrate biological information for guided feature selection.

Main Results:

• The proposed Bayesian biclustering method demonstrated robust performance across various data types.
• Feature selection was effectively guided by incorporating biological information.
• Comparative analyses showed superior performance against existing biclustering methods on multi-omics datasets.

Conclusions:

• The novel Bayesian biclustering method offers a flexible and powerful approach for analyzing diverse omics data.
• Integrating biological information significantly enhances feature selection and analytical performance.
• This method provides a valuable tool for uncovering complex patterns in multi-omics studies.