A multiscale model to substitute constitutive equations used in computational fluid dynamics (CFD) is proposed in this work, capable of linking the macroscopical rheology of viscoelastic coolants with the molecular composition of the polymeric solutions. The modelling framework is based on molecular dynamics (MD) and many-body dissipative particle dynamics (mDPD) providing a physics-informed data input for artificial neural networks (ANNs) in the form of stress-strain rate correlations. Aqueous and non-aqueous solvent-based solutions mixed with surfactants and/or polymers of different chemistries and morphologies will be examined. Solvents comprise water, polyalphaolefins and esters while anionic, cationic, non-ionic and zwitterionic surfactants will be dissolved in the base matrices at concentrations up to 2%. In addition, linear, branched, cross-linked and other shaped-polymer chain morphologies of different chemistries will be simulated in the mDPD regime. The thermodynamic (mass density, cohesive energy density, etc.) and transport (viscosity, diffusivity, etc.) properties obtained from atomistic MD simulations are subsequently utilised to parametrise the mDPD models. At a second step, the mDPD model is used to simulate five different flows, namely, shear flow, uniaxial extensional flow, biaxial extensional flow, 2-dimensional planar flow and wall-bounded flow using non-equilibrium MD (NEMD) method. The combined dataset from MD and mDPD provides stress-strain rate correlations along with other properties of the fluids for strain rates from 1 s-¹ to 10¹¹ s-¹. The dataset is used to train an ANN architecture embedded with invariance properties correlating the strain rate tensor, thermodynamic and relaxation properties to the produced stress tensor for benchmark flows.

Protein adsorption onto material surfaces is a key factor in both the biocompatibility of medical devices and drug stability in pharmaceutical containers. Understanding the physicochemical factors governing these interactions is essential for optimizing biomaterial design and minimizing drug loss due to sorption. In this study, we employed multi-scale molecular modeling to investigate protein adsorption onto polyvinyl chloride (PVC) surfaces, both plasticized and non-plasticized.

At the atomistic scale, molecular dynamics (MD) simulations revealed that human serum albumin (HSA) exhibits the highest affinity (among the human serum proteins) for PVC, with adsorption driven primarily by van der Waals forces and water-mediated hydrogen bonding, while maintaining its structural integrity. Free energy calculations confirmed a thermodynamically favorable adsorption process.

To extend our understanding to larger systems, we utilized the Martini 3 coarse-grained model to study insulin monomer and hexamer interactions with PVC surfaces containing di(2-ethylhexyl) terephthalate (DEHT) and tris(2-ethylhexyl) trimellitate (TOTM) plasticizers. This approach allowed us to capture protein adsorption dynamics on complex surfaces, evaluate the role of plasticizers, and achieve improved sampling of the potential of mean force.

Our findings provide new insights into the interplay between proteins, polymers, and plasticizers, highlighting their implications for biocompatibility and drug stability. This multi-scale approach bridges the gap between experimental and computational studies, offering a comprehensive framework for designing optimized biomaterials and drug delivery systems.

Since the water permeation mechanisms of polyamide membranes are highly dependent on the micro-structure of polyamide membranes, a deeper understanding of the formation process of them is necessary for the optimization of the membrane performance. As most simulation works construct the polyamide membranes in an ideal way, the interfacial diffusion of monomers, which is crucial for the formation of polyamide membranes, is usually ignored. To address this issue, we mimic the experimental conditions to develop an atomic model of a highly crosslinked polyamide membrane by conducting molecular dynamics simulations. Via tuning the concentration and molar ratio of monomers, the diffusion of monomers and its influence on the subsequent reaction are altered, and the final membrane structure consequently changed. Simulation results reveal that increasing the trimesoyl chloride concentration results in thicker membranes with a reduced specific surface area and consequently decreased water permeance. On the other hand, increasing the m-phenylenediamine concentration will accelerate the reaction rate and reduce the final crosslinking degree. A deeper understanding of the mechanism of polyamide-membrane formation is unveiled in this work, which can aid in the design of high-performance polyamide membranes in the future.

Surface modification presents an appealing approach for tailoring the characteristics of polymer membranes. However, the lack of predictive relationships between modification strategies and membrane performance remains challenging. Addressing this, we explore the potential of data-driven approaches, specifically machine learning. Our study employs machine learning methods on datasets comprising performance metrics of modified membranes to predict parameters such as pure water permeability and zeta potential for membranes modified with novel substances.
Machine-learning models were used to predict novel modified membranes. Additionally, the new membranes were prepared. The predicted membrane properties (pure water permeance and zeta potential) were compared to the measured values.
In this study, machine learning methods were first used to identify the importance of a substance’s chemical structures and process parameters on the resulting membrane properties. The importance can help to understand which of the modification parameters affected the properties of the modified membranes, and, therefore, new modification strategies can be chosen based on these data.
Additionally, membranes modified with new substances were predicted. After the preparation of these new membranes, the predicted results were compared to the measured properties. The predictions from both models were excellent, especially for interpolated values of experimental conditions for the electron beam-based modification approach, e.g., irradiation dose and concentration. The predictions were done in a short computational time and with satisfying accuracy.
In conclusion, the application of machine learning in membrane modification is a promising tool for accelerating the development of membranes with improved performance and for saving time/costs during the development process

Despite the modern advances in the available computational resources, the length and time scales of the physical systems that can be studied in full atomic detail, via molecular simulations, are still limited. To overcome such limitations, methods based on hierarchical multi-scale modelling have been developed.

In recent years, deep learning (DL) has emerged as a powerful tool for addressing the complexities of representing such high-dimensional functions. This breakthrough offers an unprecedented opportunity to revisit and enhance the theoretical foundations of various scientific fields, develop innovative methodologies, and solve problems that were previously too complex for traditional approaches.

Here, we explore several of these challenges within the domain of multi-scale modeling. We propose DL-based techniques that significantly enhance simulation accuracy and efficiency, surpassing the capabilities of conventional methods. We focus on molecular modeling, employing theoretical methods and computational techniques to simulate and study molecular behavior, from small chemical systems to large multi-component molecular systems and material assemblies. In this work, we introduce a series of DL-enhanced multi-scale models designed to overcome these limitations and achieve more accurate and efficient simulations. These models focus on predicting the mechanical properties of polymer nanocomposite systems [1] and restoring atomic detail in coarse-grained multi-component macromolecules [2,3].

The primary objective of this work is to systematically demonstrate that these methods eliminate the complexity of using complicated physical based relations, while managing to reduce the computation time by several orders of magnitude, thereby showcasing the transformative potential of DL in advancing multi-scale modeling.

Star polyelectrolytes (SPE) are macromolecules featuring a central core from which multiple charged polymer arms extend, offering unique structural and electrostatic properties. Building on our previous work2 on the glass transition of charged spherical micelles composed of SPEs, we extend our investigation to explore the structural conformation and phase behaviour of highly charged SPE solutions with oppositely charged linear polyelectrolyte (LPE) additives across varied concentrations, charge ratios and LPE lengths. Through molecular dynamics simulations, we demonstrate that electrostatic interactions, packing effects, and chain connectivity fundamentally dictate SPE structure. SPEs maintain a fully extended conformation at dilute concentrations until the overlap threshold is reached, beyond which enhanced intermolecular interactions induce contraction. The introduction of LPEs further modulates this behaviour, as even small quantities can completely neutralise the effective charge of the SPEs, leading to the formation of bridging chains that connect separate SPE cores. Longer LPEs exhibit a higher probability of bridging, promoting local clustering, and inducing phase separation at high charge ratios into distinct polyelectrolyte-rich and supernatant phases. Conversely, mixtures containing shorter LPEs remain more homogeneous with less pronounced bridging. Additionally, a structural reentrance phenomenon is observed, where SPE core–core correlations initially intensify with concentration before diminishing past the overlap point. Experimental rheological measurements corroborate these findings, confirming that increased LPE content drives phase separation. These insights provide a foundation for tailoring polyelectrolyte blends in applications requiring controlled complexation and responsiveness, such as drug delivery and developing advanced functional materials.