مدل‌سازي فرآيند تنظيم ژن با شبكه‌هاي بيزي پوياي ناهمگن

Gene regulatory networks (GRNs) derived from time-series gene expression data deman‎d advanced computational approaches to effectively capture the dynamic an‎d non-stationary nature of biological processes. Traditional Dynamic Bayesian Network (DBN) models often rely on the assumption of network homogeneity, which diminish their accuracy when applied to real-world datasets characterized by temporally variable interactions. This research examines an improved framework based on a non-homogeneous Dynamic Bayesian Network (NH-DBN) that incorporates a multiple change-point process alongside node-specific adaptive sampling methods. A novel strategy utilizing Manhattan distance is employed to enhance the detection of change points by measuring distributional disparities between adjacent temporal segments. Also, a Markov chain Monte Carlo (MCMC) approach with a node-specific particle filter is introduced to improve model convergence an‎d refine parent set selec‎tion. The study also presents a Bayesian seemingly unrelated regression model for the simultaneous inference of gene regulatory networks an‎d residual dependencies. This model allows predictor selec‎tion fitted to each response while simultaneously accounting for residual correlations within a Bayesian network framework. Such features reduce the risk of model overcomplexity. A specialized MCMC-based procedure is formulated for tasks involving parameter estimation, dependency learning, an‎d structural inference. Empirical analyses conducted on Saccharomyces cerevisiae (yeast) an‎d RAF signaling pathway datasets indicate that the proposed method surpasses existing homogeneous an‎d non-homogeneous models in both network reconstruction accuracy an‎d convergence speed. These advancements contribute to a more robust computational basis for dynamic modeling of gene regulatory networks across varying temporal scenarios.

4