Thesis defense of Maria Kelidou

This thesis presents an automated framework for the coarse-grained parameterisation of small molecules using mixed-variable particle swarm optimisation (PSO). Coarse-grained molecular models extend the accessible time and length scales of molecular dynamics simulations, but their parameterisation often requires substantial manual effort and expert knowledge. To address this challenge, a systematic optimisation workflow is developed within the CGCompiler framework, combining molecular dynamics simulations with mixed-variable PSO to efficiently explore mixed discrete-continuous parameter spaces.
The thesis reviews the theoretical foundations of particle swarm optimisation, molecular dynamics simulation methods, and coarse-grained modelling approaches relevant to small molecule parameterisation. Within the workflow, coarse-grained model parameters are iteratively optimised by comparing simulation results against reference data obtained from experiments and atomistic simulations using suitable target observables.
The methodology is first applied to dopamine and serotonin, two biologically relevant small molecules for which no existing coarse-grained Martini models were available. To benchmark the automated approach against manual parameterisation strategies, additional parameterisations are performed for two small molecules with existing manually developed Martini models. The presented framework contributes towards more reproducible and automated strategies for coarsegrained small molecule parameterisation.