Partially non-homogeneous dynamic Bayesian networks based on Bayesian regression models with partitioned design...

Area of Science:

• Systems Biology
• Computational Biology
• Bioinformatics

Background:

• Non-homogeneous dynamic Bayesian networks (NH-DBNs) are widely used for inferring cellular networks from time series data.
• Biological experiments often involve multiple conditions where only specific network parameters vary.
• Existing models may not efficiently handle condition-specific parameter changes and non-equidistant time points.

Purpose of the Study:

• To develop a novel partially NH-DBN model capable of handling condition-specific parameter variations in biological networks.
• To integrate a Gaussian process-based method for analyzing non-equidistant time series data.
• To improve the accuracy of network reconstruction in systems biology.

Main Methods:

• Proposed a partially NH-DBN framework utilizing Bayesian hierarchical regression with partitioned design matrices.
• Implemented a Gaussian process-based approach to address non-equidistant time series measurements.
• Applied the model to semi-quantitative immunoblot timecourse data of the mammalian target of rapamycin complex 1 (mTORC1) signaling pathway.

Main Results:

• The new model demonstrated improved network reconstruction accuracy on synthetic and yeast gene expression data.
• Successfully reconstructed the topologies of the circadian clock network in Arabidopsis thaliana.
• Inferred network structures for the mTORC1 signaling pathway align with existing biological literature.

Conclusions:

• The developed partially NH-DBN model offers a robust approach for analyzing condition-specific biological networks.
• The integration of Gaussian processes effectively handles non-equidistant time series data, a common challenge in systems biology.
• The model's application provides biologically consistent insights into complex regulatory networks.