Outcome-guided Bayesian clustering for disease subtype discovery using high-dimensional transcriptomic data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Complex diseases exhibit significant heterogeneity, necessitating disease subtyping for precision medicine.
• Omics data aids in identifying disease subtypes, but traditional clustering may lack clinical relevance.
• Clinical data offers potential to guide omics-based subtyping for meaningful discoveries.

Purpose of the Study:

• To develop an outcome-guided Bayesian clustering method integrating omics and clinical data.
• To discover clinically meaningful disease subtypes and associated genes.
• To enable precision medicine through robust disease subtyping.

Main Methods:

• Developed Guided Bayesian Clustering (GuidedBayesianClustering), a full Bayesian approach.
• Utilized Gaussian mixture models for sample clustering and spike-and-slab priors for gene selection.
• Incorporated clinical outcome variables using mixture model priors and employed Gibbs sampling for efficiency.

Main Results:

• Simultaneously achieved disease subtype discovery, feature selection, and outcome-guided subtyping.
• Demonstrated superior performance through simulations and real-world data applications (breast cancer, Alzheimer's disease).
• An R package is available on GitHub for broader application.

Conclusions:

• Guided Bayesian Clustering effectively integrates omics and clinical data for discovering clinically relevant disease subtypes.
• This method advances precision medicine by providing more meaningful patient stratification.
• The publicly available R package facilitates the adoption of this advanced subtyping technique.