Dynamic particle swarm optimization of biomolecular simulation parameters with flexible objective functions

Abstract Molecular simulations are a powerful tool to complement and interpret ambiguous experimental data on biomolecules obtain structural models. Such data-assisted often rely parameters, the choice of which is highly non-trivial crucial performance. The key challenge weighting information with respect underlying physical model. We introduce FLAPS, self-adapting variant dynamic particle swarm optimization, overcome this parameter selection problem. FLAPS suited for optimization composite objective functions that depend both parameters additional, priori unknown substantially influence search-space topology. These learned at runtime, yielding dynamically evolving iteratively refined As practical example, we show how can be used find functional small-angle X-ray scattering-guided protein simulations.