An improved Bayesian network method for reconstructing gene regulatory network based on candidate auto selection.

Area of Science:

• Systems biology
• Bioinformatics
• Computational biology

Background:

• Reconstructing gene regulatory networks (GRNs) from gene expression data is crucial for understanding complex biological mechanisms.
• Current methods like Bayesian networks (BNs) are powerful but computationally expensive, while mutual information methods are fast but lack directionality and have high false-positive rates.
• Addressing these limitations is essential for advancing systems biology and bioinformatics.

Purpose of the Study:

• To develop a novel algorithm that accelerates Bayesian network learning for GRN inference.
• To improve the efficiency and accuracy of computational approaches for analyzing gene regulatory mechanisms.

Main Methods:

• Proposed the Candidate Auto Selection (CAS) algorithm, which leverages mutual information and breakpoint detection to restrict the search space for BN structure learning.
• Introduced two methods based on CAS: CAS + G (globally optimal greedy search) and CAS + L (local learning).

Main Results:

• The CAS algorithm effectively reduces the search space for Bayesian networks by identifying neighbor candidates for each node.
• CAS + G demonstrated superior performance compared to state-of-the-art methods in inferring GRNs from simulation data.
• CAS + L achieved significantly faster learning speeds than existing methods with minimal loss in accuracy.

Conclusions:

• CAS-based methods substantially decrease the computational complexity associated with Bayesian network analysis.
• The proposed CAS methods offer a more suitable and efficient approach for GRN inference in systems biology and bioinformatics.