A method for efficient Bayesian optimization of self-assembly systems from scattering data.

Area of Science:

• Biophysics
• Computational Biology
• Systems Biology

Background:

• Molecular self-assembly into functional complexes is fundamental to cellular processes.
• Viral capsids serve as a model system for studying complex macromolecular assembly.
• Previous work utilized stochastic simulation and static light scattering (SLS) to infer assembly kinetics.

Purpose of the Study:

• To develop and validate a novel Bayesian optimization framework for analyzing molecular assembly.
• To enhance the inference of kinetic rate parameters using richer experimental data.
• To apply the framework to both stochastic and deterministic simulation models.

Main Methods:

• Introduced a Bayesian optimization framework with multi-Gaussian process model regression.
• Extended prior work to incorporate small-angle X-ray/neutron scattering (SAXS/SANS) data.
• Validated the method using synthetic data from viral protein structures and differential equation models.

Main Results:

• The new framework effectively constrains assembly trajectories and infers kinetic parameters.
• SAXS/SANS data provided richer insights compared to SLS.
• The approach was successfully applied to computationally cheaper differential equation models.

Conclusions:

• Presented a flexible approach for global optimization of complex dynamic models.
• Outperformed state-of-the-art black box solvers for stochastic viral capsid assembly.
• The method demonstrates broad applicability to general stochastic optimization problems.