BIC-LP: A Hybrid Higher-Order Dynamic Bayesian Network Score Function for Gene Regulatory Network Reconstruction.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Gene regulatory network (GRN) reconstruction is crucial in bioinformatics.
• Dynamic Bayesian Networks (DBNs) are common but face challenges from network complexity and data noise, leading to inaccurate edges.
• Existing methods like ScoreLasso have limitations, including first-order assumptions.

Purpose of the Study:

• To propose an integrated model and a novel hybrid higher-order DBN score function, BIC-LP, for improved GRN reconstruction.
• To address the limitations of existing methods by incorporating higher-order models and reducing information loss.

Main Methods:

• Developed an integrated model combining higher-order DBN, higher-order Lasso linear regression, and Pearson correlation models.
• Constructed the BIC-LP score function by augmenting the classical BIC score with terms from Lasso regression and Pearson correlation coefficients.
• Evaluated the performance of BIC-LP against existing Bayesian score functions and state-of-the-art GRN reconstruction methods.

Main Results:

• The proposed BIC-LP score function effectively captures more information from datasets compared to existing methods.
• BIC-LP demonstrates superior performance in GRN reconstruction by reducing false positive edges while preserving true positive edges.
• Experimental results validate the enhanced accuracy and reliability of the BIC-LP approach.

Conclusions:

• BIC-LP offers a significant advancement in GRN reconstruction by providing a more accurate and robust scoring function.
• The integration of higher-order modeling and complementary statistical measures enhances the ability to decipher complex gene regulatory interactions.
• This method holds promise for more precise biological network analysis and discovery.