Nanoplastics interact continuously with circulating immune cells, yet how particle size and exposure complexity shape immune transcriptional organization under physiological flow conditions remains poorly understood. Here, controlled microfluidic exposure was combined with single-cell RNA sequencing to investigate the effects of size-defined polystyrene nanoplastics (PSNPs; 40 nm, 200 nm, and 40 + 200 nm) on primary human peripheral blood mononuclear cells (PBMCs) under dynamic flow conditions. Across immune populations, PSNP exposure induced a conserved ribosome-associated and RNA-regulatory transcriptional program, indicating a shared intracellular adaptive response. Monocytes displayed the strongest transcriptional remodeling, characterized by coordinated modulation of ribosome-associated, metabolic, and inflammatory signaling pathways in a size-dependent manner. Exposure to 40 nm PSNPs negatively enriched (suppressed) mitochondrial metabolic pathways, whereas 200 nm PSNPs enriched inflammatory signaling programs. Combined exposure induced concurrent metabolic and inflammatory pathway engagement without evidence of major immune topology disruption or discrete inflammatory state transitions. In contrast, adaptive immune cells exhibited comparatively modest and lineage-preserving transcriptional modulation. Together, these findings demonstrate that nanoplastic size and exposure complexity shape coordinated immunometabolic adaptation in human immune cells under physiologically relevant flow conditions and establish a framework for studying dynamic material–immune interactions at single-cell resolution.​

The FitzHugh–Nagumo (FHN) equation in one dimension is solved in this paper using an improved physics-informed neural network (PINN) approach. Examining test problems with known analytical solutions and the explicit finite difference method (EFDM) allowed for the demonstration of the PINN’s effectiveness. Our study presents an improved PINN formulation tailored to the FitzHugh–Nagumo reaction–diffusion system. The proposed framework is efficiently designed, validated, and systematically optimized, demonstrating that a careful balance among model complexity, collocation density, and training strategy enables high accuracy within limited computational time. Despite the very strong agreement that both methods provide, we have demonstrated that the PINN results exhibit a closer agreement with the analytical solutions for Test Problem 1, whereas the EFDM yielded more accurate results for Test Problem 2. This study is crucial for evaluating the PINN’s performance in solving the FHN equation and its application to nonlinear processes like pulse propagation in optical fibers, drug delivery, neural behavior, geophysical fluid dynamics, and long-wave propagation in oceans, highlighting the potential of PINNs for complex systems. Numerical models for this class of nonlinear partial differential equations (PDEs) may be developed by existing and future model creators of a wide range of various nonlinear physical processes in the physical and engineering sectors using the concepts of the solution methods employed in this study.

We investigate oxygen diffusion in the soil in one dimension by finite differences and the physics-informed neural network. Solving the diffusion equation by either method determines the oxygen concentration profiles inside the soil column at various times. However, while respecting specified Dirichlet and Neumann boundary conditions, the concentration profiles at certain times become negative, which is non-physical per se. We can resolve this situation in finite differences by proclaiming these negative concentration values as zero during the time-stepping scheme. In the case of PINN, we propose an innovative solution with a custom loss function, tailored to avoid such non-physical behavior. Two types of Dirichlet boundary conditions are investigated. The first is constant, and the second one periodically changes, with a period of 24 hours. We demonstrate that the PINN with a customized loss is effective and accurate. The proposed approach to circumvent non-physical solution areas demonstrates promise for application to various analogous problems.

This review presents the key aspects and development directions of materials informatics, emphasizing the role of artificial intelligence (AI) and machine learning (ML) in materials science research. The objective is to provide a comprehensive overview of materials informatics tools, workflows, and case studies, particularly aimed at experimental researchers unfamiliar with AI frameworks. Basic concepts are introduced and traditional modelling methods compared to AI/ML-assisted models. Existing material models serve as a foundation for advanced modelling and simulations aimed at reducing the time required for characterisation and discovery, with physics-based models gaining importance in the development of AI-supported surrogate models. This review also covers currently available resources, including: (i) software for solving complex mathematical equations and material modelling; (ii) web-based platforms and tools designed for both expert and non-expert users; and (iii) materials data repositories, prioritising standardisation. Case examples involving materials with architectured macro-, micro-, and nano-porosity are reviewed across three material types: metal-organic frameworks (MOFs), electrospun PVDF piezoelectrics, and 3D printed mechanical metamaterials. Traditional computational models offer interpretability and physical consistency, AI/ML excels in speed and complexity handling but may lack transparency. Hybrid models combining both approaches show excellent results in prediction, simulation, and optimisation, offering both speed and interpretability. Progress depends on modular, interoperable AI systems, standardised FAIR data, and cross-disciplinary collaboration. Addressing data quality and integration challenges will resolve issues related to metadata gaps, semantic ontologies, and data infrastructures, especially for small datasets and unlock transformative advances in fields like nanocomposites, MOFs, and adaptive materials.

This review article presents key aspects of applying Smoothed Particle Hydrodynamics (SPH) to model fluid filtration through FDM 3D-printed copper-based filters (80% Cu, 20% PLA composite filament) with microporosity. Conservation laws in continuum dynamics form the foundation of meshless SPH modeling. The complexity of modeling 3D printed copper-based filters lies in capturing the irregular shapes of microporosity. We present an initial SPH model that provides velocity and pressure distributions through a round pore, demonstrating better accuracy than the FEM model. Such SPH approach can accurately capture water flow parameters through complex porous structures. Unlike FEM, which relies on predefined finite elements, SPH is a Lagrangian method based on the distribution of virtual particles and periodic boundary conditions, making it well-suited for modelling irregular geometries. Combined with 3D printing, SPH can support the design of optimal structures for water filtration.

BACKGROUND AND OBJECTIVE:In accordance with the latest aspirations in the field of bioengineering, there is a need to create a web accessible, but powerful cloud computational platform that combines datasets and multiscale models related to bone modeling, cancer, cardiovascular diseases and tissue engineering. The SGABU platform may become a powerful information system for research and education that can integrate data, extract information, and facilitate knowledge exchange with the goal of creating and developing appropriate computing pipelines to provide accurate and comprehensive biological information from the molecular to organ level.METHODS:The datasets integrated into the platform are obtained from experimental and/or clinical studies and are mainly in tabular or image file format, including metadata. The implementation of multiscale models, is an ambitious effort of the platform to capture phenomena at different length scales, described using partial and ordinary differential equations, which are solved numerically on complex geometries with the use of the finite element method. The majority of the SGABU platform's simulation pipelines are provided as Common Workflow Language (CWL) workflows. Each of them requires creating a CWL implementation on the backend and a user-friendly interface using standard web technologies. Platform is available at https://sgabu-test.unic.kg.ac.rs/login.RESULTS:The main dashboard of the SGABU platform is divided into sections for each field of research, each one of which includes a subsection of datasets and multiscale models. The datasets can be presented in a simple form as tabular data, or using technologies such as Plotly.js for 2D plot interactivity, Kitware Paraview Glance for 3D view. Regarding the models, the usage of Docker containerization for packing the individual tools and CWL orchestration for describing inputs with validation forms and outputs with tabular views for output visualization, interactive diagrams, 3D views and animations.CONCLUSIONS:In practice, the structure of SGABU platform means that any of the integrated workflows can work equally well on any other bioengineering platform. The key advantage of the SGABU platform over similar efforts is its versatility offered with the use of modern, modular, and extensible technology for various levels of architecture.

This study employs a novel physics-informed neural network (PINN) approach, the standard explicit finite difference method (EFDM) and unconditionally positivity preserving FDM to tackle the one-dimensional Sine–Gordon equation (SGE). Two test problems with known analytical solutions are investigated to demonstrate the effectiveness of these techniques. While the three employed approaches demonstrate strong agreement, our analysis reveals that the EFDM results are in the best agreement with the analytical solutions. Given the consistent agreement between the numerical results from the EFDM, unconditionally positivity preserving FDM and PINN approach and the analytical solutions, all three methods are recommended as competitive options. The solution techniques employed in this study can be a valuable asset for present and future model developers engaged in various nonlinear physical wave phenomena, such as propagation of solitons in optical fibers.

The paper presents the GeNNsem (Genetic algorithm ANNs ensemble) software framework for the simultaneous optimization of individual neural networks and building their optimal ensemble. The proposed framework employs a genetic algorithm to search for suitable architectures and hyperparameters of the individual neural networks to maximize the weighted sum of accuracy and diversity in their predictions. The optimal ensemble consists of networks with low errors but diverse predictions, resulting in a more generalized model. The scalability of the proposed framework is ensured by utilizing micro-services and Kubernetes batching orchestration. GeNNsem has been evaluated on two regression benchmark problems and compared with related machine learning techniques. The proposed approach exhibited supremacy over other ensemble approaches and individual neural networks in all common regression modeling metrics. Real-world use-case experiments in the domain of hydro-informatics have further demonstrated the main advantages of GeNNsem: requires the least training sessions for individual models when optimizing an ensemble; networks in an ensemble are generally simple due to the regularization provided by a trivial initial population and custom genetic operators; execution times are reduced by two orders of magnitude as a result of parallelization.

Biophysical muscle models based on sliding filament and cross-bridge theory are called Huxley-type muscle models. The method of characteristics is typically used to solve Huxley's muscle contraction equation, which describes the distribution of attached myosin heads to the actin-binding sites, called cross-bridges. Once this equation is solved, we can determine the generated force and the stiffness of the muscle fibers, which can then be used at the macro level during finite element analysis. In our paper, we present alternative approaches to finding an approximate solution of Huxley's muscle contraction equation using neural networks. In one approach, we collect the data from simulations and train multilayer perceptrons to predict probabilities of cross-bridge formation based on the available actin site positions, time, activation, current and previous stretch. In another approach, besides using the data, we also inform the neural network with Huxley's equation, thus improving the generalization of the neural network's predictions.

We employed a novel physics-informed neural networks (PINN) to tackle (1+1) dimensional Sine-Gordon equation (SGE). A test problem with known analytical solution is numerically solved to demonstrate the effectiveness of the PINN, and compared to standard explicit finite difference method (EFDM) and Chen-Charpentier et al.’s finite difference method (CCFDM). Although the three methods used show good agreement, our research shows that the EFDM results accord with the analytical solution the best. Given the consistent agreement between the numerical results obtained by all three methods and the analytical solution, all three methods are recommended as competitive options. The methods employed in this study can be a valuable asset for present and future model developers engaged in various nonlinear physical wave phenomena, such as solitons evolution or soliton interaction as well as optical pulses propagation in nonlinear media, including optical fibers.

Huxley-type muscle models offer a physiologically grounded description of cardiac contraction but remain computationally prohibitive for large-scale, multi-scale simulations. This article reviews surrogate modeling strategies that alleviate these costs for ventricular biomechanics, with emphasis on data-driven (RNN/TCN/GRU) and physics-informed (PINN) formulations and their coupling to finite-element solvers. The data-driven approach utilizes deep neural networks trained on numerical simulation data to replicate the behavior of the Huxley model while significantly reducing processing costs. The physics-informed approach approximates solutions to Huxley’s muscle contraction equation, which governs cross-bridge dynamics and force generation. By predicting the probability of myosin-actin interactions, this method enables direct calculation of stress and stiffness for finite element simulations. The coupling of these surrogate models with finite element computational frameworks allows for faster and more scalable simulations. Our goal is to provide a consolidated reference and actionable guidance for selecting and implementing surrogate approaches for Huxley-type muscle simulations.

There is a need to develop an integrated computational platform that will contain both datasets and multiscale models related to bone (modelling), cancer, cardiovascular diseases, and tissue engineering. The SGABU platform is a robust information system capable of data integration, information extraction, and knowledge exchange, with the goal of designing and developing suitable computing pipelines to give accurate and adequate biological information from the patient's molecular to organ level. Datasets integrated into the platform are directly obtained from experimental and/or clinical studies and are mostly in tabular or image file format. Multiscale models range from models that can be described using partial or ordinary differential equations, to complex models that use finite element modelling. The majority of the SGABU platform's simulation modules are built as Common Workflow Language workflows. This implies creating a CWL implementation on the Functional Engine Service backend and creating an acceptable User Interface. The key advantage of SGABU platform is the utilization of new, contemporary, modular, and unique technology for various levels of architecture.

Supervised deep learning requires a huge amount of reference data, which is often difficult and expensive to obtain. Domain adaptation helps with this problem—labelled data from one dataset should help in learning on another unlabelled or scarcely labelled dataset. In remote sensing, where a variety of sensors produce images of different modalities and with different numbers of channels, it would be very beneficial to develop heterogeneous domain adaptation methods that are able to work between domains that come from different input spaces. However, this challenging problem is rarely addressed, the majority of existing heterogeneous domain adaptation work does not use raw image-data, or they rely on translation from one domain to the other, therefore ignoring domain-invariant feature extraction approaches. This article proposes novel approaches for heterogeneous image domain adaptation for both the semi-supervised and unsupervised settings. These are based on extracting domain invariant features using deep adversarial learning. For the unsupervised domain adaptation case, the impact of pseudo-labelling is also investigated. We evaluate on two heterogeneous remote sensing datasets, one being RGB, and the other multispectral, for the task of land-cover patch classification, and also on a standard computer vision benchmark of RGB-depth map object classification. The results show that the proposed domain invariant approach consistently outperforms the competing methods based on image-to-image/feature translation, in both remote sensing and in a standard computer vision problem.

When simulating various physical phenomena, the law of the phenomenon is often known in advance, in the form of a partial differential equation, that needs to be solved. Numerical methods, such as the finite element method, have been developed over decades, and these methods approximate the solution to the partial differential equation. However, these methods can be computationally demanding. On the oth- er hand, neural networks, can provide predictions that approximate the given partial differential equation. Neural networks are computationally more efficient than numerical methods, but they often face issues of generalization and consequently problems with solution accuracy. Insufficient generalization, among other things, can result from data collected from numerical simulations. In the last few years, physics-informed neural networks are being developed, for which it’s not necessary to gather data from simulations. These networks use automatic differentiation and during training, they minimize the residuals of the partial differ- ential equation, its initial, and boundary conditions. After training, these neural networks can be used as a replacement for traditional numerical solvers.

The Burgers’ equation is solved using the explicit finite difference method (EFDM) and physics-informed neural networks (PINN). We compare our numerical results, obtained using the EFDM and PINN for three test problems with various initial conditions and Dirichlet boundary conditions, with the analytical solutions, and, while both approaches yield very good agreement, the EFDM results are more closely aligned with the analytical solutions. Since there is good agreement between all of the numerical findings from the EFDM, PINN, and analytical solutions, both approaches are competitive and deserving of recommendation. The conclusions that are provided are significant for simulating a variety of nonlinear physical phenomena, such as those that occur in flood waves in rivers, chromatography, gas dynamics, and traffic flow. Additionally, the concepts of the solution techniques used in this study may be applied to the development of numerical models for this class of nonlinear partial differential equations by present and future model developers of a wide range of diverse nonlinear physical processes.

Biophysical muscle models, also known as Huxley-type models, are appropriate for simulating non-uniform and unsteady contractions. Large-scale simulations can be more challenging to use because this type of model can be computationally intensive. The method of characteristics is typically used to solve Huxley’s muscle equation, which describes the distribution of connected myosin heads to the actin-binding sites. Once this equation is solved, we can determine the generated force and the stiffness of the muscle fibers, which may then be employed in the macro-level simulations of finite element analysis. In our paper, we developed a physics-informed surrogate model that functions similarly to the original Huxley muscle model but uses a lot less computational resources in order to enable more effective use of the Huxley muscle model.

Supervised deep learning relies heavily on the existence of a huge amount of labelled data, which in many cases is difficult to obtain. Domain adaptation deals with this problem by learning on a labelled dataset and applying that knowledge to another, unlabelled or scarcely labelled dataset, with a related but different probability distribution. Heterogeneous domain adaptation is an especially challenging area where domains lie in different input spaces. These methods are very interesting for the field of remote sensing (and indeed computer vision in general), where a variety of sensors are used, capturing images of different modalities, different spatial and spectral resolutions, and where labelling is a very expensive process. With two heterogeneous domains, however, unsupervised domain adaptation is difficult to perform, and class-flipping is frequent. At least a small amount of labelled data is therefore necessary in the target domain in many cases. This work proposes loosening the label requirement by labelling the target domain with must-link and cannot-link constraints instead of class labels. Our method Constrained-HIDA, based on constraints, contrastive loss, and learning domain invariant features, shows that a significant performance improvement can be achieved by using a very small number of constraints. This demonstrates that a reduced amount of information, in the form of constraints, is as effective as giving class labels. Moreover, this paper shows the benefits of interactive supervision—assigning constraints to the samples from classes that are known to be prone to flipping can further reduce the necessary amount of constraints.

Physics-Informed Neural Networks (PINNs) are artificial neural networks that encode Partial Differential Equations (PDEs) as an integral component of the ML model. PINNs are successfully used nowadays to solve PDEs, fractional equations, and integral–differential equations, including direct and inverse problems. Just as in the case of other kinds of artificial neural networks, the architecture, including the number and sizes of layers, activation functions, and other hyperparameters can significantly influence the network performance. Despite the serious work in this field, there are still no clear directions on how to choose an optimal network architecture in a consistent manner. In practice, expertise is required, with a significant number of manual trial and error cycles. In this paper, we propose PINN/GA (PINN/Genetic Algorithm), a fully automatic design of a PINN by an evolutionary strategy with specially tailored operators of selection, crossover, and mutation, adapted for deep neural network architecture and hyperparameter search. The PINN/GA strategy starts from the population of simple PINNs, adding new layers only if it brings clear accuracy benefits, keeping PINNs in the population as simple as possible. Since the examination of dozens of neural networks through the evolutionary process implies enormous computational costs, it employs a scalable computational design based on containers and Kubernetes batching orchestration. To demonstrate the potential of the proposed approach, we chose two non-trivial direct problems. The first is 1D Stefan transient model with time-dependent Dirichlet boundary conditions, describing the melting process, and the second is the Helmholtz wave equation over a 2D square domain. The authors found that PINNs accuracy gradually improves throughout the evolutionary process, exhibiting better performance and stability than parallel random search and Hyperopt Tree of Parzen Estimators, while keeping the network design reasonably simple.

The paper analyzes how the Byzantine law influenced The Town Law of Novo Brdo, which represents the second part of Despot Stefan Lazarevic?s Novo Brdo Legal Code of 1412. A possible connection between the town law of Novo Brdo and certain provisions of The Syntagma of Matthew Blastares and one of the privileges that the town of Ioannina received from the Byzantine Emperor Andronikos II in 1319 is suggested. Accordingly, it is assumed that certain provisions of the Town Law of Novo Brdo could have been formulated during the reign of Emperor Stefan Dusan.

The investigation of the bandwidth in multimode graded-index microstructured polymer optical fiber (GI mPOF) with a solid core is proposed using a modal diffusion approach. For a variety of launch radial offsets of multimode GI mPOF, bandwidth is reported by numerically solving the time-dependent power flow equation (TD PFE) using the explicit finite difference method (EFDM) and physics-informed neural networks (PINN). The decline in bandwidth with fiber length becomes slower at fiber lengths close to the coupling length Lc at which an equilibrium mode distribution (EMD) is attained, showing that mode coupling enhances bandwidth at longer fiber lengths. As fiber length is increased, bandwidth approaches complete independence from radial offset, suggesting the steady-state distribution (SSD) has been reached. We compare multimode GI mPOF performance in terms of bandwidth with that of traditional multimode GI POFs made of the same material. Higher bandwidth performance and quicker bandwidth improvement are displayed by the GI mPOF. To enhance fiber performance in GI mPOF links, such a fiber char-acterization can be used.