A Unified Bayesian Framework for Bi-overlapping-Clustering Multi-omics Data via Sparse Matrix Factorization.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Modern sequencing technologies generate vast multi-omics data, offering insights into functional genomes.
• Diverse data types from different sequencing modalities complicate statistical modeling.
• Existing methods often require specialized approaches for each data type.

Purpose of the Study:

• To propose a unified framework for analyzing multi-omics data using Bayesian nonparametric matrix factorization.
• To infer overlapping bi-clusters within integrated multi-omics datasets.
• To develop a method that adaptively handles diverse data types.

Main Methods:

• Bayesian nonparametric matrix factorization
• Adaptive discretization of observations into common latent states
• Hierarchical construction of cluster structures
• Application to single-cell RNA-seq, single-cell ATAC-seq, bulk RNA-seq, and DNA methylation data

Main Results:

• The proposed method successfully infers overlapping bi-clusters across different omics data types.
• The framework adaptively discretizes diverse observations into shared latent states.
• The Bayesian nonparametric approach automatically determines the optimal number of clusters.
• Analysis of multiple datasets revealed biologically relevant findings consistent with literature.

Conclusions:

• The unified framework provides a powerful approach for integrated multi-omics data analysis.
• The method effectively handles data heterogeneity and identifies complex cluster structures.
• This work advances the quantitative exploration of functional genomes using multi-omics data.