Automatic multi-objective optimization of coarse-grained lipid force fields using SwarmCG

Abstract

The development of coarse-grained (CG) molecular models typically requires a time-consuming iterative tuning of parameters in order to have the approximated CG models behave correctly and consistently with, e.g., available higher-resolution simulation data and/or experimental observables. Automatic data-driven approaches are increasingly used to develop accurate models for molecular dynamics simulations. However, the parameters obtained via such automatic methods often make use of specifically designed interaction potentials and are typically poorly transferable to molecular systems or conditions other than those used for training them. Using a multi-objective approach in combination with an automatic optimization engine (SwarmCG), here, we show that it is possible to optimize CG models that are also transferable, obtaining optimized CG force fields (FFs). As a proof of concept, here, we use lipids for which we can avail reference experimental data (area per lipid and bilayer thickness) and reliable atomistic simulations to guide the optimization. Once the resolution of the CG models (mapping) is set as an input, SwarmCG optimizes the parameters of the CG lipid models iteratively and simultaneously against higher-resolution simulations (bottom-up) and experimental data (top-down references). Including different types of lipid bilayers in the training set in a parallel optimization guarantees the transferability of the optimized lipid FF parameters. We demonstrate that SwarmCG can reach satisfactory agreement with experimental data for different resolution CG FFs. We also obtain stimulating insights into the precision-resolution balance of the FFs. The approach is general and can be effectively used to develop new FFs and to improve the existing ones.

Two decades of Martini: Better beads, broader scope

“…For certain application areas, specific CG force fields optimized for the task at hand can be a better choice—examples include protein folding studies with the UNRES force field, 19 prediction of nucleotide structure and hybridization with oxDNA, 18 studying molecular fluids with SAFT‐γ, 428 or simulation of polymer dynamics either with generic models 429,430 or with bottom‐up constructed (structure‐based) CG models 431–433 . Further advancement of machine learning 434,435 will certainly contribute to this development, either by producing target force fields on demand, or in the process of integrating data from various resources to optimize existing ones, including Martini 104 . Encouraging is also to see the efforts of other CG force field developers to extend their models toward broader application ranges, providing more future opportunities to cross‐validate predictions from Martini.…”

“…Some of these focus on generating bonded potentials, such as the pyCGtool 100 and Swarm‐CG, 101 whereas others target the challenge of mapping an underlying chemical structure to its CG Martini representation like autoMartini 102 and the graph‐based cg_param algorithm 103 . Recently, the multi‐objective based Swarm‐CG algorithm has also been used as a tool to improve the nonbonded interactions of Martini lipids 104 …”

Bottom-up Coarse-Graining: Principles and Perspectives

“…This also has the added benefit of ideally making CGSchNet transferable across system composition and accurately modeling the solvation environment for biomolecules in a novel manner with ISSNet . We note that graph neural network architectures and graph-based approaches have additionally found widespread use in other molecular tasks such as the selection of CG mapping operators and automatic sampling of atomistic configurations. , The CGnet architecture has also been adapted to only consider many-body interactions up to a specified order. For example, it is shown that even 5-body interactions notably improve the quality of the resulting CG model when studying a small protein …”

Biomimicking Extracellular Vesicles with Fully Artificial Ones: A Rational Design of EV-BIOMIMETICS toward Effective Theranostic Tools in Nanomedicine

“…In this perspective, machine learning approaches have been recently considered with the aim to accelerate the development of accurate coarse-grained molecular models. 71 It is of prominent importance, in the development these molecular systems, to avail on experimental data (like geometric parameters such as area per lipid and/or bilayer thickness) and of reliable all-atom force fields. Examples were reported demonstrating the ability of advanced CG models to simulate lipid−lipid interaction, self-assembly into lipid bilayer, formation of vesicles, and vesicle fusion using different model lipids.…”