Constraint-Based, Score-Based and Hybrid Algorithms to Construct Bayesian Gene Networks in the Bovine Transcriptome.

Area of Science:

• Genomics and Bioinformatics
• Reproductive Biology
• Computational Biology

Background:

• Bayesian gene networks model causal relationships and incorporate prior knowledge for biological inferences.
• Progesterone significantly impacts bovine endometrial tissue following conception, influencing gene expression.
• Understanding gene regulatory networks in the bovine uterus is crucial for complex trait expression.

Purpose of the Study:

• To construct and compare Bayesian gene networks for genes expressed in the bovine uterus.
• To evaluate the efficacy of six different algorithms in identifying hub genes.
• To elucidate pathways influenced by progesterone in bovine uterine tissue during early pregnancy.

Main Methods:

• Analysis of Gene Expression Omnibus (GEO) dataset GSE33030 using six algorithms (grow-shrink, max-min parent children, tabu search, hill-climbing, max-min hill-climbing, restricted maximum).
• Categorization of algorithms into constraint-based, score-based, and hybrid approaches.
• Gene network parameter estimation using the bnlearn R package.

Main Results:

• The tabu search algorithm identified the highest network parameters, including gene degree, Markov blankets, neighborhood size, and branching factor.
• Hybrid and constraint-based algorithms shared the highest number of hub genes (e.g., PRODH, SPDEF, MMD2, SEMA3E, SLC27A6, ACTG2).
• Functional annotation revealed hub genes are involved in extracellular pathways, metabolism, protein structure, and post-translational processes.

Conclusions:

• Different Bayesian network algorithms and topological parameters effectively identify hub genes in the bovine uterus.
• The tabu search algorithm demonstrated superior performance in network construction.
• Identified hub genes provide insights into progesterone's role in uterine function and gene regulatory networks during early pregnancy.