Computational Model Research Articles

From Simulations to Inference: Using Machine Learning to Tune Patient-Specific Finite-Element Models of the Middle Ear Towards Objective Diagnosis.

Computational models, particularly finite-element (FE) models, are essential for interpreting experimental data and predicting system behavior, especially when direct measurements are limited. A major challenge in tuning these models is the large number of parameters involved. Traditional methods, such as one-by-one sensitivity analyses, are time-consuming, subjective, and often return only a single set of parameter values, focusing on reproducing averaged data rather than capturing the full variability of experimental measurements. In this study, we applied simulation-based inference (SBI) using neural posterior estimation (NPE) to tune an FE model of the human middle ear. The training dataset consisted of 10,000 FE simulations of stapes velocity, ear-canal (EC) input impedance, and absorbance, paired with seven FE parameter values randomly sampled within plausible ranges. The neural network learned the association between parameters and simulation outcomes, returning the probability distribution of parameter values that can reproduce experimental data. Our approach successfully identified parameter sets that reproduced three experimental datasets simultaneously. By accounting for experimental noise and variability during training, the method provided a probability distribution of parameters, representing all valid combinations that could fit the data, rather than tuning to averaged values. The network demonstrated robustness to noise and exhibited an efficient learning curve due to the large training dataset. SBI offers an objective alternative to laborious sensitivity analyses, providing probability distributions for each parameter and uncovering interactions between them. This method can be applied to any biological FE model, and we demonstrated its effectiveness using a middle-ear model. Importantly, it holds promise for objective differential diagnosis of conductive hearing loss by providing insight into the mechanical properties of the middle ear.

Realistic photon-number resolution in Gaussian boson sampling

Gaussian boson sampling (GBS) is a model of nonuniversal quantum computation that claims to demonstrate quantum supremacy with current technologies. This model entails sampling photocounting events from a multimode Gaussian state at the outputs of a linear interferometer. In this scheme, collision events—those with more than one photon for each mode—are infrequent. However, they are still used for validation purposes. Therefore, the limitation of realistic detectors to perfectly resolve adjacent photon numbers becomes pivotal. We derive a photocounting probability distribution in GBS schemes which is applicable for use with general detectors and photocounting techniques. This probability distribution is expressed in terms of functionals of the field-quadrature covariance matrix, e.g., Hafnian and Torontonian in the well-known special cases of photon-number-resolving and on-off detectors, respectively. Based on our results, we consider a GBS validation technique involving detectors with realistic photon-number resolution. Published by the American Physical Society 2024

Ranking Magnetic Colloid Performance for Magnetic Particle Imaging and Magnetic Particle Hyperthermia

AbstractMagnetic particle imaging (MPI) is an emerging modality that can address longstanding technological challenges encountered with magnetic particle hyperthermia (MPH) cancer therapy. MPI is a tracer technology compatible with MPH for which magnetic nanoparticles (MNPs) provide signal for MPI and heat for MPH. Identifying whether a specific MNP formulation is suitable for both modalities is essential for clinical implementation. Current models predict that functional requirements of each modality impose conflicting demands on nanoparticle magnetic properties. This objective here is to develop a measurement and ranking scheme based on end‐use performance to streamline evaluation of candidate MNP formulations. The measured MPI point‐spread function (PSF) and specific loss power (SLP) is combined to generate a single numerical value for comparison on a relative ranking scale, or figure of merit (FoM). 12 aqueous iron‐containing formulations are evaluated, including FDA‐approved (parenteral) iron‐containing colloids. MNPs with high (Synomag‐D70: 123.4), medium (Synomag‐D50: 63.2), and low (NanoXact: 0.147) FoM values are selected for in vivo validation of the selection scheme in subcutaneous 4T1 tumors. Results demonstrate that the proposed ranking accurately assessed the relative performance of MNPs for MPI and MPH. Data demonstrated that image quality and tumor temperature rise increased with FoM ranking, validating predictions. It isshown that the MPI signal correlated with MNP concentration in tissue. Computational heat transfer models anchored on tumor MPI data harmonized with experimental results to within an average of 2 °C when MNP content estimated from MPI data is included. Computational studies emphasized the importance of post‐injection MNP quantitation and MPI spatial resolution.

Biomedical relation extraction method based on ensemble learning and attention mechanism

BackgroundRelation extraction (RE) plays a crucial role in biomedical research as it is essential for uncovering complex semantic relationships between entities in textual data. Given the significance of RE in biomedical informatics and the increasing volume of literature, there is an urgent need for advanced computational models capable of accurately and efficiently extracting these relationships on a large scale.ResultsThis paper proposes a novel approach, SARE, combining ensemble learning Stacking and attention mechanisms to enhance the performance of biomedical relation extraction. By leveraging multiple pre-trained models, SARE demonstrates improved adaptability and robustness across diverse domains. The attention mechanisms enable the model to capture and utilize key information in the text more accurately. SARE achieved performance improvements of 4.8, 8.7, and 0.8 percentage points on the PPI, DDI, and ChemProt datasets, respectively, compared to the original BERT variant and the domain-specific PubMedBERT model.ConclusionsSARE offers a promising solution for improving the accuracy and efficiency of relation extraction tasks in biomedical research, facilitating advancements in biomedical informatics. The results suggest that combining ensemble learning with attention mechanisms is effective for extracting complex relationships from biomedical texts. Our code and data are publicly available at: https://github.com/GS233/Biomedical.

A percolation-type criticality threshold controls immune protein coating of surfaces.

When a material enters the body, it is immediately attacked by hundreds of proteins, organized into complex networks of binding interactions and reactions. How do such complex systems interact with a material, "deciding" whether to attack? We focus on the "complement" system of ∼40 blood proteins that bind microbes, nanoparticles, and medical devices, initiating inflammation. We show a sharp threshold for complement activation upon varying a fundamental material parameter, the surface density of potential complement attachment points. This sharp threshold manifests at scales spanning single nanoparticles to macroscale pathologies, shown here for diverse engineered and living materials. Computational models show these behaviors arise from a minimal subnetwork of complement, manifesting percolation-type critical transitions in the complement response. This criticality switch explains the "decision" of a complex signaling network to interact with a material, and elucidates the evolution and engineering of materials interacting with the body.

Intrinsic dynamic shapes responses to external stimulation in the human brain.

Sensory stimulation of the brain reverberates in its recurrent neuronal networks. However, current computational models of brain activity do not separate immediate sensory responses from intrinsic recurrent dynamics. We apply a vector-autoregressive model with external input (VARX), combining the concepts of "functional connectivity" and "encoding models", to intracranial recordings in humans. We find that the recurrent connectivity during rest is largely unaltered during movie watching. The intrinsic recurrent dynamic enhances and prolongs the neural responses to scene cuts, eye movements, and sounds. Failing to account for these exogenous inputs, leads to spurious connections in the intrinsic "connectivity". The model shows that an external stimulus can reduce intrinsic noise. It also shows that sensory areas have mostly outward, whereas higher-order brain areas mostly incoming connections. We conclude that the response to an external audiovisual stimulus can largely be attributed to the intrinsic dynamic of the brain, already observed during rest.

Binding mode of cyclic chemerin-9 peptide and chemerinS157 protein at CMKLR.

The chemokine-like receptor 1 (CMKLR1) is activated by the adipokine and chemoattractant protein chemerin. Cryo-EM structures of chemerin-9-CMKLR1-Gi have been published, where chemerin-9 is the nonapeptide of the C terminus of chemerinS157. Chemerin-9 is as active as the full-length protein in Ca2+-release but shows differences in equilibrium read-outs. An equally potent cyclic chemerin-9 variant (cC9) was reported previously. Now, we have built a computational model of CMKLR1 to investigate the binding mode of cC9 and chemerinS157 in comparison to chemerin-9. Differences were investigated using CMKLR1 variants. Double-mutant cycle analysis identified CMKLR1-F2.53 as the relevant position for Phe8-binding of cC9. Energy contribution revealed slight differences in Phe8-binding to CMKLR1-F2.53 and space for larger residues. This was confirmed as the chemerin-9 variant with 1-naphthyl-L-alanine at position 8 showed a 4-fold increased potency of 2 nM (pEC50=8.6±0.15). While chemerin-9 and cC9 share their interactions at the CMKLR1, chemerinS157 tolerates most mutations of CMKLR1 in the deep binding site. The computational model of chemerinS157 suggests a β-sheet interaction between the N-terminal CMKLR1-segment I25VVL28 and the β-sheet D108KVLGRLVH116 of ChemS157, which was confirmed experimentally. Our data expand the knowledge by identifying the binding mode of chemerinS157 and cC9 at CMKLR1 facilitating future structure-based drug design.

A cotton waste reinforced composite for automotive applications: development and thermal characterization

Fibre-reinforced composites are largely used in automobiles and building materials in various forms. Thermal insulation and acoustic characteristics are being mainly considered in building construction and automotive applications to meet overall comfort conditions as well as fulfil energy efficiency requirements. It is critical to look for new and sustainable approaches to increasing the thermal characteristics of the composites. Integration of various wastes is one of the most sustainable and efficient methods for increasing the thermal characteristics of composites. The present work is aimed to investigate the thermal insulation performance of cotton waste-reinforced composites using both epoxy and modified epoxy using the C-THERM thermal conductivity analyzer according to ASTM C518 standards. A computational model of randomly dispersed epoxy matrix interphase with cotton fibers is developed and validated with experimental results. It was observed that the thermal conductivity is higher when the combination of 40% flake cotton waste and 60% matrix material is used, both for epoxy and modified epoxy. The thermal conductivity for epoxy composites is 0.5715 W/mK, while for modified epoxy composites it is 0.4935 W/mK. The results indicate that modified epoxy composites exhibit better thermal insulation properties, as they have lower thermal conductivity values compared to epoxy composites. The modified epoxy composite is considered the best for thermal insulation, can be utilised in various automotive applications, such as partitioning panels, automotive panels, brake pads, and similar applications.

An evaluation of square footing response on lime-treated geotextile-reinforced silty sand: contrasting experimental and computational approaches

Improving soil strength and reducing the anticipated settlement and construction cost is a great paradox for civil as well as geotechnical engineers. In this paper, these aspects and other suitable types of ground improvement are discussed based on the principles of using geosynthetics for soil reinforcement. A series of load-settlement tests were also performed to compare strength and settlement of the silty sand reinforced with lime and one layer of geotextile. The study finds the maximum insertions of geotextile at 0.2D (3.0 cm) beneath the square footing base, and the lime percentage of 5.0% increases the UBC substantially. The UBC of lime-treated and geotextile-reinforced silty sand was to an optimum of 1,360 kN/m2 that has shown an enhancement of 258% compared to that of untreated and unreinforced silty sand that is approximately 380 kN/m2. Furthermore, comparative analysis between two ANN models was performed to provide improved estimate of the UBC, namely artificial neural network (ANN) and extreme learning machine (ELM). The developed computational models were then compared with experiment data, which proved that such models are more economical and effective than the expensive and time consuming conventional techniques. Consequently, based on the results, it was further validated that ELM possesses better generalization capability compared to ANN for predictive efficiency and thereby proves the efficiency of the model in estimating the ultimate bearing capacity of square footings incorporated with geotextile and lime-treated silty sand. This places the ELM model as a useful tool in the initial conceptual as well as the design for improvement steps of soil reinforcement.

Excited-State Decay and Photolysis of O-Nitrophenol after Proton Transfer. II: A Theoretical Investigation in the Microsolvated Atmospheric Environment.

Changes in atmospheric humidity affect the number of water molecules surrounding o-nitrophenol (ONP), creating an anisotropic chemical environment. It, in turn, influences the photodynamic behaviors of ONP, differing from those observed in the gas phase and in solution. Recently, we explored the excited-state decay and the generation of the hydroxyl (OH) radical before proton transfer of ONP in the microsolvated environment using the MS-CASPT2//CASSCF approach. As is well known, ONP is capable of converting to its aci-nitro isomer (aciONP) via an excited-state intramolecular proton transfer (ESIPT) process. In the present work, the photoinduced dynamics of aciONP, which can lead to an OH radical and nitrous acid (HONO), was studied using the same computational model. Our calculations demonstrated that increasing the number of water molecules affects the molecular geometries, particularly the key bond lengths and dihedral angles of the HONO group, while also reducing the relative energies of minima and intersections. Moreover, we identified two distinct types of minimum structures: one that retains the intramolecular hydrogen bond and the other that breaks the hydrogen bond with the H atom flipping outward. The latter structure, compared with the former, has a different electronic-state character and facilitates intersystem crossing processes. Subsequently, two major excited-state decay paths were proposed: (PATH I) ESIPT → S1 → S1S0 → S0; (PATH II) ESIPT → S1 → S1-2 → S1T1 → T1 → S0T1 → S0. Furthermore, the T1 state has a relatively long lifetime, allowing for the formation of the OH radical and HONO, and the corresponding energy barriers decrease as the number of water molecules increases. These theoretical findings provide valuable insights into the photodynamics of aciONP in the microsolvated atmospheric environment.

Mechanisms of memory-supporting neuronal dynamics in hippocampal area CA3.

Hippocampal CA3 is central to memory formation and retrieval. Although various network mechanisms have been proposed, direct evidence is lacking. Using intracellular Vm recordings and optogenetic manipulations in behaving mice, we found that CA3 place-field activity is produced by a symmetric form of behavioral timescale synaptic plasticity (BTSP) at recurrent synapses among CA3 pyramidal neurons but not at synapses from the dentate gyrus (DG). Additional manipulations revealed that excitatory input from the entorhinal cortex (EC) but not the DG was required to update place cell activity based on the animal's movement. These data were captured by a computational model that used BTSP and an external updating input to produce attractor dynamics under online learning conditions. Theoretical analyses further highlight the superior memory storage capacity of such networks, especially when dealing with correlated input patterns. This evidence elucidates the cellular and circuit mechanisms of learning and memory formation in the hippocampus.

Functional excitation-inhibition ratio indicates near-critical oscillations across frequencies

Abstract The concept of excitation/inhibition (E/I) balance plays an important role in understanding brain function in health and disease. We recently introduced an algorithm to determine a functional E/I ratio based on the critical brain dynamics that emerge in neuronal networks balancing between order and disorder. Little, however, is known about the frequency specificity of E/I regulation and how to measure it. Here, we optimized the algorithm for measuring functional excitation-inhibition ratio (fE/I) in narrow frequency ranges and validated it on a computational model of critical oscillations and EEG data. In the computational model, we confirmed that fE/I discriminated E/I connectivity differences across a wide range of frequencies (1–150 Hz). Twin EEG data revealed significant genetic influences on fE/I across frequencies, whereas contrasting eyes-open and -closed EEG indicated functional changes of fE/I restricted to a subset of alpha and beta oscillations and brain regions. We propose that assessing fE/I with finer frequency resolution will prove useful for understanding the functional role of E/I regulation in a spectrally refined fashion in health and disease.

Damage Evolution Mechanism of Railway Wagon Bogie Adapter 1035 Steel and Damage Parameter Calibration Based on Gursone-Tvergaarde-Needleman Model.

As a critical component of a train, the railway wagon bogie adapter has higher quality requirements. During the forging process, external loads can induce voids, cracks, and other defects in the forging, thereby reducing its service life. Hence, studying the damage behavior of the forging material, specifically AISI 1035 steel, becomes crucial. This study involved obtaining stress-strain curves for AISI 1035 steel through uniaxial tensile tests at temperatures of 900 °C, 1000 °C, and 1100 °C, with strain rates of 0.1 s-1, 1 s-1, and 10 s-1. Subsequently, SEM was used to observe samples at various deformation stages. The damage parameters, q1, q2 and q3 in the GTN model "a computational model used to analyze and simulate material damage which can effectively capture the damage behavior of materials under different loading conditions" were then calibrated using the Ramberg-Osgood model and stress-strain curve fitting. Image Pro Plus software v11.1 quantified the sample porosity as f0, fn, fc and fF. A finite element model was established to simulate the tensile behavior of the AISI 1035 steel samples. By comparing the damage parameters of f0, fn, fc and fF obtained by the finite element method and experimental method, the validity of the damage parameters obtained by the finite element inverse method could be verified.

Computational fluid-particle dynamic model guiding bioengineered magnetic nanomedicine for personalized brain-targeted drug delivery

Computational fluid-particle dynamic model guiding bioengineered magnetic nanomedicine for personalized brain-targeted drug delivery

Computational Fluid Dynamics Modelling of Hydrogen Production via Water Splitting in Oxygen Membrane Reactors.

The utilization of oxygen transport membranes enables the production of high-purity hydrogen by the thermal decomposition of water below 1000 °C. This process is based on a chemical potential gradient across the membrane, which is usually achieved by introducing a reducing gas. Computational fluid dynamics (CFD) can be used to model reactors based on this concept. In this study, a modelling approach for water splitting is presented in which oxygen transport through the membrane acts as the rate-determining process for the overall reaction. This transport step is implemented in the CFD simulation. Both gas compartments are modelled in the simulations. Hydrogen and methane are used as reducing gases. The model is validated using experimental data from the literature and compared with a simplified perfect mixing modelling approach. Although the main focus of this work is to propose an approach to implement the water splitting in CFD simulations, a simulation study was conducted to exemplify how CFD modelling can be utilized in design optimization. Simplified 2-dimensional and rotational symmetric reactor geometries were compared. This study shows that a parallel overflow of the membrane in an elongated reactor is advantageous, as this reduces the back diffusion of the reaction products, which increases the mean driving force for oxygen transport through the membrane.

Comparative Analysis of Machine Learning Techniques for Identifying Multiple Force Systems from Accelerometer Measurements.

The knowledge of the forces acting on a structure enables, among many other factors, assessments of whether the component's useful life is compromised by the current machine condition. In many cases, a direct measurement of those forces becomes unfeasible, and an inverse problem must be solved. Among the solutions developed, machine learning techniques have stood out as powerful predictive tools increasingly applied to engineering problem-solving. This study evaluates the ability of different machine learning models to identify parameters of multi-force systems from accelerometer measurements. The models were assessed according to their prediction potential based on correlation coefficient (R2), mean relative error (MRE), and processing time. A computational numerical model using the finite element method was generated and validated by vibration measurements performed using accelerometers in the laboratory. A robust database created by the response surface methodology in conjunction with Design of Experiment (DOE) was used for the evaluation of the ability of machine learning models to predict the position, frequency, magnitude, and number of forces acting on a structure. Among the six machine learning models evaluated, k-NN was able to predict with a 0.013% error, and Random Forests showed a maximum error of 0.2%. The innovation of this study lies in the application of the proposed method for identifying parameters of multi-force systems.

Fluid mediated communication among flexible micro-posts in chemically reactive solutions.

Communication in biological systems typically involves enzymatic reactions that occur within fluids confined between the soft, elastic walls of bio-channels and chambers. Through the inherent transformation of chemical to mechanical energy, the fluids can be driven to flow within the confined domains. Through fluid-structure interactions, the confining walls in turn are deformed by and affect this fluid flow. Imbuing synthetic materials with analogous feedback among chemo-mechanical, hydrodynamic and fluid-structure interactions could enable materials to perform self-driven communication and self-regulation. Herein, we develop computational models to determine how chemo-hydro-mechanical feedback affects interactions in biomimetic arrays of chemically active and passive micro-posts anchored in fluid-filled chambers. Once activated, the enzymatic reactions trigger the latter feedback, which generates a surprising variety of long-range, cooperative motion, including self-oscillations and non-reciprocal interactions, which are vital for propagating coherent, directional signals over net distances in fluids. In particular, the array propagates a distinct message; each post interprets the message; and the system responds with a specific mode of organized, collective behavior. This level of autonomous remote control is relatively rare in synthetic systems, particularly as this system operates without external electronics or power sources and only requires the addition of chemical reactants to function.

Invented models – relating students’ constructions of computational models to their learning gains

ABSTRACT Does inventing representations in open-ended computational environments contradict the goal of acquiring canonical scientific knowledge? We analyze the computational models of electric conductors that N = 35 eight-grade students built with a NetLogo-based microworld. Students used the microworld to sketch structures that represent models of electric conductors and then observed the resulting electric current. We use data from classroom discourse, computer log files and questionnaires, to identify different ways through which students used the platform, and how their engagement in construction, explains the variance in their learning gains. We found differences in the number and types of models that students constructed in the microworld: while most students built the models that were intended by the instructors, some added unconventional doodling models, that deviated from the intended structures. We found that students who built more models made larger gains in conceptual knowledge about electric current. However, we also found that students who constructed more unconventional models, produced lower learning gains on items that assessed the particle-level mechanism of electrons in conductors.

The meaning of attention control.

Attention control has been proposed as an ability construct that explains individual differences in fluid intelligence. Evaluating this hypothesis is complicated by a lack of clarity in the definition of attention control. Here, I propose a definition of attention control, based on experimental research and computational models of what guides attention, and how cognitive processes are controlled. Attention is the selection of mental representations for prioritized processing, and the ability to control attention is the ability to prioritize those representations that are relevant for the person's current goal, thereby enabling them to think and act in accordance with their intentions. This definition can be used to identify appropriate and less appropriate ways to measure individual differences in attention control. An analysis of various approaches to measurement reveals that the current practice of measuring attention control leaves room for improvement. Aligning our psychometric measurements with a clear, theoretically grounded concept of attention control can lead to more valid measures of that construct. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Three-dimensional anisotropic unified continuum model for simulating the healing of damaged soft biological tissues.

The soft biological tissues have the ability to heal and self-repair after damage or injury. During the healing process, damaged tissues are replaced by newly produced undamaged tissue to restore homeostasis. Computational modeling serves as an effective tool for simulating the healing process and understanding the underlying mechanisms. In previous work, we developed the first unified continuum damage model for the healing of soft biological tissues. However, the initial theory lacked generalizability to more realistic scenarios and applicability to biomechanical problems due to the simplicity of the isotropic constitutive model and two-dimensional simulations. Therefore, we further improve our approach by developing a three-dimensional anisotropic unified healing model to address more realistic challenges. By using the Holzapfel-Gasser-Ogden model as the hyperelastic term, the influence of the collagen fibers is considered and the reorientation of fibers in healing is simulated. Three numerical examples related to hypertension, aneurysm, and restenosis of the atherosclerotic artery after balloon angioplasty are presented to demonstrate the effectiveness of the proposed model. By comparing numerical solutions and reference solutions, we demonstrate the ability of the proposed model in simulating long-term tissue healing process and analyze the impact of anisotropic terms.