Kernel-Based Particle Filtering for Scalable Inference in Partially Observed Boolean Dynamical Systems.

Area of Science:

• Systems Biology
• Computational Biology
• Network Security

Background:

• Partially Observed Boolean Dynamical Systems (POBDS) are effective for modeling gene dynamics and network security.
• Existing inference methods for POBDS face scalability issues due to high computational costs.
• Efficient parameter inference is crucial for understanding complex biological and security systems.

Purpose of the Study:

• To develop a scalable inference method for Partially Observed Boolean Dynamical Systems (POBDS).
• To address the computational challenges in likelihood evaluation and parameter space exploration for POBDS.
• To improve the efficiency of modeling complex biological and security networks using POBDS.

Main Methods:

• A kernel-based particle filtering approach is proposed.
• Gaussian processes (GPs) are utilized to approximate the likelihood function efficiently.
• Bayesian optimization is employed to guide parameter search, balancing exploration and exploitation.

Main Results:

• The proposed method significantly improves scalability for POBDS inference.
• Gaussian processes effectively represent uncertain likelihood evaluations.
• Bayesian optimization efficiently identifies optimal parameters, reducing computational burden.

Conclusions:

• The kernel-based particle filtering method offers a scalable solution for POBDS inference.
• This approach enhances the applicability of POBDS in systems biology and network security.
• The method was successfully validated on mammalian cell-cycle and T-cell leukemia networks.