Bayesian Topology Inference of Regulatory Networks under Partial Observability.

Area of Science:

• Systems biology and genomics, focusing on network inference.
• Computational biology and bioinformatics, utilizing advanced statistical methods.

Background:

• Biological systems like microbial communities and gene regulatory networks (GRNs) involve complex interactions with noisy data.
• Reconstructing network topology is challenging due to scale, high dimensionality, and noise, limiting current inference techniques.
• Existing methods often face issues with scalability, interpretability, and overfitting in large biological datasets.

Purpose of the Study:

• To develop an efficient and scalable Bayesian topology optimization framework for inferring regulatory networks.
• To model biological networks as partially-observed Boolean dynamical systems (POBDS).
• To overcome the limitations of existing inference techniques for large and complex biological systems.

Main Methods:

• Proposed a Bayesian topology optimization framework for network inference.
• Combined the Boolean Kalman Filter (BKF) as an optimal estimator for POBDS.
• Employed Bayesian optimization with Gaussian Process regression and a topology-inspired kernel function to model the log-likelihood.

Main Results:

• Demonstrated superior performance in numerical experiments.
• Accurately inferred topology in the p53-MDM2 network with 8 and 16 unknown regulations.
• Identified the correct topology for the mammalian cell cycle network with 10 unknown regulations, showing lower error and faster convergence.

Conclusions:

• The proposed Bayesian framework offers an efficient and scalable solution for inferring regulatory network topology.
• The method effectively handles noisy and high-dimensional biological data.
• This approach advances the analysis of complex biological systems, enabling better understanding of gene regulation and cellular dynamics.