A fully Bayesian latent variable model for integrative clustering analysis of multi-type omics data.

Area of Science:

• Computational biology
• Bioinformatics
• Cancer research

Background:

• Precision medicine relies on identifying clinically relevant tumor subtypes and omics signatures.
• Large-scale projects like The Cancer Genome Atlas (TCGA) provide extensive multi-omics data.
• Existing methods for integrative clustering of multi-omics data lack computational efficiency.

Purpose of the Study:

• To develop a computationally efficient, fully Bayesian latent variable method (iClusterBayes) for integrative clustering of multi-omics data.
• To jointly model continuous and discrete omics data for tumor subtyping and omics feature identification.
• To improve upon existing integrative clustering methods like iClusterPlus.

Main Methods:

• Developed a fully Bayesian latent variable model (iClusterBayes).
• Employs latent variables for joint dimension reduction of multiple omics datasets.
• Utilizes Bayesian variable selection to identify omics features driving sample clustering.

Main Results:

• iClusterBayes demonstrates superior statistical inference and computational speed compared to iClusterPlus.
• Successfully identified clinically meaningful tumor subtypes and driver omics features in TCGA and simulated datasets.
• The method effectively clusters tumor samples in a latent variable space.

Conclusions:

• iClusterBayes provides an efficient and statistically robust approach for integrative multi-omics analysis.
• The method facilitates the discovery of novel tumor subtypes and biomarkers for precision oncology.
• This tool advances cancer translational research by enabling deeper insights from complex genomic data.