Hidden Markov induced Dynamic Bayesian Network for recovering time evolving gene regulatory networks

Area of Science:

• Computational Systems Biology
• Bioinformatics
• Genomics

Background:

• Dynamic Bayesian Networks (DBNs) are standard for gene regulatory network inference from time-series data.
• The stationarity assumption in DBNs limits their ability to model evolving biological systems.
• Existing non-stationary DBN methods face challenges with computational time, accuracy, and parameter tuning.

Purpose of the Study:

• To develop a novel non-stationary DBN model that accurately captures time-evolving gene regulatory networks.
• To improve the computational efficiency and reduce reliance on parameter settings for network inference.
• To enhance the accuracy of gene regulatory network reconstruction from biological time-series data.

Main Methods:

• Proposed a Hidden Markov Model with Dynamic Bayesian Networks (HMDBN) by extending hidden nodes into DBNs.
• Developed an improved structural Expectation-Maximization (EM) algorithm for efficient HMDBN learning.
• Derived a generalized Bayesian Information Criterion under non-stationarity (BWBIC) to improve accuracy and reduce overfitting.
• Derived re-estimation formulas for model parameters to eliminate the need for manual settings.

Main Results:

• The HMDBN model demonstrated significantly improved computational efficiency by reducing the search space.
• The BWBIC criterion led to substantially higher reconstruction accuracy and reduced model overfitting.
• The method achieved stably high prediction accuracy on both synthetic and real biological datasets.
• The proposed approach outperformed state-of-the-art methods in terms of accuracy and efficiency.

Conclusions:

• The novel HMDBN model effectively addresses the limitations of stationary DBNs for gene regulatory network inference.
• The developed methods provide a more accurate, efficient, and robust approach to analyzing time-evolving biological networks.
• This work offers a valuable tool for computational systems biology, particularly for understanding dynamic biological processes.