Gene regulatory network inference based on a nonhomogeneous dynamic Bayesian network model with an improved Markov...

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Dynamic Bayesian Networks (DBNs) struggle with non-stationary gene expression data.
• Existing models face challenges in network reconstruction accuracy and model convergence.

Purpose of the Study:

• To enhance DBNs for non-stationary gene expression data.
• To improve network reconstruction accuracy and model convergence speed.

Main Methods:

• Developed an MD-birth move using Manhattan distance for multi-change point processes.
• Implemented a node-dependent particle filtering-based Markov chain Monte Carlo sampling method.
• Proposed the FInal Convergence Dynamic Bayesian Network (FC-DBN) model.

Main Results:

• The MD-birth move increases the rationality of the multi-change point process.
• Node-dependent particle filtering improves sampling efficiency for gene regulatory network reconstruction.
• FC-DBN demonstrates superior network reconstruction accuracy and faster convergence compared to existing models on Saccharomyces cerevisiae and RAF datasets.

Conclusions:

• The proposed FC-DBN model effectively addresses limitations in modeling non-stationary gene expression data.
• The novel methods enhance the performance of dynamic Bayesian networks for gene regulatory network inference.