H-CORE: enabling genome-scale Bayesian analysis of biological systems without prior knowledge

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Bayesian networks model probabilistic relationships between data entities using directed acyclic graphs (DAGs).
• Previous Bayesian network learning methods face scalability limitations due to large search spaces, restricting analysis to fewer entities or requiring prior knowledge.
• High-throughput biological data necessitates scalable methods for inferring complex biological relationships.

Purpose of the Study:

• To introduce the hierarchical clustering and order restriction (H-CORE) method for learning large-scale Bayesian networks.
• To address the scalability problem in Bayesian network learning for genome-scale analysis.
• To enable Bayesian network analysis of biological systems without relying on additional biological knowledge.

Main Methods:

• The H-CORE method utilizes hierarchical clustering to group biological entities.
• Edge directions are restricted between clusters to manage the search space during Bayesian network structure learning.
• The method was evaluated using simulations and applied to gene-to-gene relationship inference in the 'Rosetta compendium' dataset.

Main Results:

• Simulations demonstrated that H-CORE is significantly faster than the sparse candidate method while maintaining comparable quality.
• H-CORE successfully inferred gene-to-gene relationships from the 'Rosetta compendium'.
• Literature mining confirmed the validity of the learned relationships, showcasing H-CORE's effectiveness.

Conclusions:

• H-CORE provides a scalable solution for learning large Bayesian networks, particularly for genome-scale biological data analysis.
• The method successfully overcomes the computational challenges associated with large search spaces in Bayesian network learning.
• H-CORE facilitates the analysis of complex biological systems without the need for pre-existing biological knowledge.