BAYESIAN SPARSE GRAPHICAL MODELS FOR CLASSIFICATION WITH APPLICATION TO PROTEIN EXPRESSION DATA.

Area of Science:

• Biomedical data science
• Computational biology
• Cancer research

Background:

• Reverse-phase protein array (RPPA) enables high-throughput protein network analysis.
• Accurate data modeling and network identification are crucial for RPPA data interpretation.
• Existing methods struggle with classifying biological samples based on protein network activation patterns.

Purpose of the Study:

• To develop a novel Bayesian sparse graphical modeling approach for RPPA data analysis.
• To improve the accuracy of biological sample classification using protein network data.
• To identify distinct biological relationships and signaling pathways in different cancer types.

Main Methods:

• Proposed a Bayesian sparse graphical model incorporating selection priors and class information.
• Developed a unified hierarchical model for integrated network and outcome analysis.
• Applied the model to RPPA data from human breast and ovarian cancer cell lines.

Main Results:

• The Bayesian model achieved higher accuracy in distinguishing between breast and ovarian cancer cell lines compared to existing models.
• Identified differential regulation of the PI3K-AKT signaling pathway between the two cancer types.
• Demonstrated the model's ability to perform posterior inference on network topologies within and between classes.

Conclusions:

• The proposed Bayesian sparse graphical model is a powerful tool for analyzing RPPA data.
• This approach enhances the understanding of protein networks and their role in cancer.
• The method offers improved accuracy in cancer classification and pathway analysis.