Quantitative utilization of prior biological knowledge in the Bayesian network modeling of gene expression data

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Gene expression data is noisy and limited for reconstructing genetic regulatory networks.
• Prior biological knowledge can enhance network reconstruction accuracy.
• Integrating multiple, diverse prior knowledge sources is beneficial due to individual limitations.

Purpose of the Study:

• To develop a novel method for quantitatively integrating multiple sources of prior biological knowledge into Bayesian Network (BN) modeling.
• To improve the accuracy and reliability of genetic regulatory network reconstruction.

Main Methods:

• Utilized Naïve Bayesian classifier to assess gene pair functional linkage likelihood from prior knowledge (PubMed cocitation, Gene Ontology annotation).
• Created a candidate network edge reservoir weighted by linkage likelihood.
• Employed Markov Chain Monte Carlo sampling for network generation and simulation.

Main Results:

• The new method, incorporating prior knowledge, increased recovery of known transcription regulations by approximately two-fold.
• False positive rates remained largely unchanged.
• Bayesian Network modeling without prior knowledge showed performance no better than random selection in some cases.

Conclusions:

• The developed statistical method effectively leverages quantitative prior biological knowledge within BN models.
• This integration significantly enhances the performance of genetic regulatory network reconstruction from gene expression data.