Bayesian disease classification using copy number data.

Area of Science:

• Genomics
• Bioinformatics
• Cancer Research

Background:

• DNA copy number variations (CNVs) are implicated in cancer development and progression.
• Array-based comparative genomic hybridization (aCGH) is a key technology for high-resolution CNV detection.
• Accurate CNV profiling is crucial for understanding cancer and developing personalized therapies.

Purpose of the Study:

• To develop and validate a statistical methodology for predicting disease states using aCGH profiles.
• To identify specific genomic regions associated with distinct cancer subtypes.
• To enable the development of molecular-based personalized cancer treatments.

Main Methods:

• A two-stage Bayesian classification model integrating hidden Markov models (HMMs) and Bayesian linear variable selection.
• Stage 1: HMMs infer underlying copy number states from aCGH data, accounting for probe dependencies.
• Stage 2: Bayesian variable selection identifies genomic regions associated with disease outcomes, conditional on copy number states.

Main Results:

• The proposed method accurately predicts disease categories using simulated datasets.
• The methodology successfully identified significant genomic regions associated with breast cancer subtypes in a real-world dataset.
• The selected genomic features serve as predictive parameters for classifying new individuals' disease status.

Conclusions:

• The developed two-stage Bayesian approach provides a robust framework for disease state prediction from aCGH data.
• This method enhances the understanding of CNVs in cancer and facilitates the discovery of biomarkers for personalized medicine.
• The findings support the utility of aCGH in advancing cancer diagnostics and therapeutic strategies.