OUTCOME-GUIDED DISEASE SUBTYPING BY GENERATIVE MODEL AND WEIGHTED JOINT LIKELIHOOD IN TRANSCRIPTOMIC APPLICATIONS

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput omics data enables molecular disease subtyping for distinct patient groups.
• Conventional clustering may fail to identify clinically relevant subtypes if irrelevant variables dominate.
• Need for methods that integrate clinical outcome information into disease subtyping.

Purpose of the Study:

• To develop novel outcome-guided disease subtyping methods using high-dimensional omics data.
• To improve the identification of clinically meaningful patient subgroups.
• To establish a new paradigm for precision medicine through outcome-associated subtyping.

Main Methods:

• Proposed two outcome-guided subtyping methods: a generative model and a weighted joint likelihood model.
• Both models link outcome association and subtyping via latent cluster labels.
• Weighted joint likelihood balances outcome association and gene expression patterns.

Main Results:

• Outcome-guided methods demonstrated superior performance in accuracy, gene selection, and outcome association.
• Validated through extensive simulations and real-world applications in lung disease and breast cancer.
• Weighted joint likelihood showed improved generalizability in independent validation.

Conclusions:

• Outcome-guided clustering offers a robust framework for molecular disease subtyping.
• These methods enable direct identification of patient subgroups with clinical relevance.
• Presents a new precision medicine approach for personalized patient stratification.