Development of large-scale molecular and nanomaterial models

Molecular simulations can access unique atomic-scale information about new materials, pharmaceuticals, and biological environments, making cost-effective predictions and aiding experimental studies. They are particularly useful for describing the mechanisms of nanoscale phenomena and the biological/inorganic interfaces. However, the computational cost of molecular simulations increases with the size of the system as well as with the model complexity, which is related to the accuracy of the simulation. This thesis aims to develop efficient large-scale molecular models that capture important structural details of the atomistic simulations. In particular, we focus on the TiO 2 -lipid interface, which forms in the living cells, exposed to TiO 2 nanomaterials, but is also relevant in the context of biomedical applications. We have studied the interface using atomistic molecular dynamics simulations and found that the characteristics of the lipid adsorption depend on the type of the TiO 2 surface, lipid headgroup composition, and the presence of cholesterol. We then derive a coarse-grained molecular model of the TiO 2 -lipid interface to enable the large-scale simulations of TiO 2 nanoparticles interacting with model cell membranes. We show that the strength of the lipid adsorption increases with the size of the nanoparticle and that a small TiO 2 nanoparticle can become partially wrapped by a lipid membrane. To improve the transferability of the coarse-grained model, we design and test an artificial neural network that learns the interactions in coarse-grained water-methanol solutions from the structural data obtained in multiple reference simulations at atomistic resolution. We show that in the studied system, the neural network learns the many-body interactions and accurately reproduces the structural properties of the solution at different concentrations. Molekylära simuleringar kan ge tillgång till unik information på atomnivå om nya material, läkemedel och biologiska miljöer, vilket gör det möjligt att göra kostnadseffektiva förutsägelser och underlätta experimentella studier. De är särskilt användbara för att beskriva mekanismerna för fenomen på nanoskala och de biologiska/oorganiska gränssnitten. Beräkningskostnaden för molekylära simuleringar ökar dock med systemets storlek såväl som med modellens komplexitet, vilket är relaterat till simuleringens noggrannhet. Den här avhandlingen syftar till att utveckla effektiva storskaliga molekylära modeller som