Bayesian Structure Learning for Graphical Models With Symmetry Constraints.

Area of Science:

• Genomics
• Systems Biology
• Bioinformatics

Background:

• PAM50 gene expression profiling is crucial for understanding breast cancer.
• Identifying gene relationships and pathways is key to elucidating disease mechanisms.
• Existing methods may not fully capture symmetrical network structures.

Purpose of the Study:

• To develop a graphical Gaussian model incorporating symmetry constraints for PAM50 gene expression data.
• To simultaneously identify gene dependency and symmetrical network structures.
• To deepen the understanding of functional similarities and gene regulatory behavior in breast cancer.

Main Methods:

• Utilized a graphical Gaussian model with equality constraints on the concentration matrix.
• Employed a birth-death Markov Chain Monte Carlo method for network structure inference.
• Incorporated a stepwise model learning strategy and posterior distribution approximation for efficiency.

Main Results:

• Successfully recovered dependency and symmetrical networks from PAM50 gene expression data.
• Demonstrated the model's ability to capture inherent symmetrical properties in gene regulatory behavior.
• Validated the approach through simulation studies and a practical application.

Conclusions:

• The proposed model effectively reveals symmetrical gene networks, offering new insights into breast cancer biology.
• Prioritizing genes based on functional similarity aids in understanding disease progression.
• This method provides a robust framework for analyzing complex gene expression data.