搜索结果：kernels

Dual-energy computed tomography (DECT) enables improved volumetric bone mineral density (vBMD) assessment by accounting for marrow alterations associated with aging, disease, and injury. However, DECT reconstruction kernels and monochromatic energy pair combinations may influence vBMD measurements and finite element model (FEM)-estimated bone stiffness. This study investigated the effects of reconstruction kernel and DECT energy pair combinations on proximal humeral vBMD and FEM-estimated stiffness in cadaveric specimens. Fourteen cadaveric shoulders from seven specimens were scanned bilaterally using DECT with a K2HPO4 calibration phantom. Images were reconstructed using standard (STD) and bone-sharpening (BONE) kernels. Simulated monochromatic images at 40, 90, and 140 keV were combined into 40/90, 40/140, and 90/140 keV energy pairs. Volumetric BMD was extracted from the humeral shaft diaphysis and anatomic neck using custom Python scripts and 3D Slicer software. Image-based FEMs were generated to estimate bone stiffness. Results were analyzed using repeated-measures analysis of variance (RM-ANOVA). In the cortical-dense humeral diaphysis, energy pair combinations had the greatest variation. Mean diaphyseal vBMD increased from 332.08 ± 102.54 mgK2HPO4/cc (40/90 keV BONE) to 406.84 ± 130.15 mgK2HPO4/cc (90/140 keV BONE), while FEM stiffness increased from 180.30 ± 47.07 kN/mm to 223.30 ± 63.91 kN/mm. Significant differences were observed across reconstruction conditions and energy pair combinations. In contrast, trabecular-rich anatomic neck regions demonstrated minimal variation in vBMD and FEM stiffness across energy pairs and kernels. These findings indicate that DECT energy pairs and reconstruction kernel substantially influence cortical bone assessments, particularly when using the 90/140 keV energy pair, while trabecular-rich regions remain comparatively unaffected.

Introduction: Detection of in-stent restenosis by cardiac CT is challenging due to blooming artifacts. The technological progress of CT scanners and especially the recent introduction of photon-counting detectors (PCDs) has led to an improvement in image quality. Several studies have analyzed the lumen visibility of coronary stents, but most studies used models which did not simulate cardiac movement. In this study we use a pulsatile heart model to simulate a heartbeat to analyze the effects of cardiac motion on image quality. Methods: Seventeen different coronary stents with an outer diameter of 3.0 mm were implanted into polyolefin tubes. The tubes were then filled with diluted contrast medium and attached to the pulsatile heart model. The stents were scanned in a third-generation dual-source CT with an energy-integrating detector (EID) and a first-generation PCD CT. Results: In motion, the mean visible stent lumen was reduced from 64.4% to 59.4% in EID CT, from 61.4% to 56.0% in PCD CT using the Bv60 kernel, and from 72.9% to 62.9% in PCD CT using the Bv72 kernel, each in standard resolution mode. Employing the ultra-high-resolution mode (UHR), stent lumen visibility was reduced from 61.3% to 57.9% with the Bv60 kernel and from 71.7% to 61.8% with the Bv72 kernel. The difference between static imaging and motion was significant in each instance (p < 0.001). Conclusions: While PCD CT and the use of sharper kernels improves the image quality in comparison with EID CT and smoother kernels, the impact of cardiac motion on the reduction in stent lumen visibility is substantial. Hence, the best image quality is achieved in patients with a normal and regular heart rate. If this is not possible to achieve, a retrospective acquisition mode should be considered.

Spodoptera frugiperda (J. E. Smith) is a major invasive pest of maize in Asia, with migration pathways extending annually in July into North China, a key maize-producing region. However, the potential utility of Bt maize for controlling this pest remains unclear. In this study, we evaluated the insecticidal activity and field control efficacy of Bt-(Cry1Ab + Cry1F) maize against the insect in North China. Bt insecticidal protein levels in Bt-(Cry1Ab + Cry1F) maize tissues followed the order: V6-8 leaves > V12 leaves > VT tassels, R4 kernels > R1 silks. Correspondingly, larval mortality was 100&#x2009;&#xb1;&#x2009;0% after leaf feeding for 7&#x2009;days and 84.49&#x2009;&#xb1;&#x2009;1.71%, 75.65&#x2009;&#xb1;&#x2009;1.50%, and 82.57&#x2009;&#xb1;&#x2009;1.08% after feeding on VT tassels, R1 silks, and R4 kernels, respectively. V5 leaf feeding significantly reduced pupal weight, pupation and emergence rates, and fecundity of 1st- and 3rd-instar S. frugiperda, while prolonging larval development. Field infestation experiments showed significantly lower larval density and leaf and plant damage rates of S. frugiperda in Bt-(Cry1Ab + Cry1F) maize than those in conventional maize fields, with larval control efficacy of 61.14-100%. Field surveys in 2023-2025 revealed no S. frugiperda larvae in Bt-(Cry1Ab + Cry1F) maize fields throughout the entire growth period, with plants exhibiting no damage. In contrast, conventional maize suffered continuous damage during the reproductive growth stage, with a peak larval population of 9.33-12.67 per 100 plants and a plant damage rate of 64-74%. This study demonstrates that Bt-(Cry1Ab + Cry1F) maize can effectively control this pest in the region and provide a green tool for area-wide management of S. frugiperda. &#xa9; 2026 Society of Chemical Industry.

(-)-Epigallocatechin has multiple bioactivities. ANR catalyzes its biosynthesis, with unclear structure and mechanism restricting enzyme modification. Here, we first screened ANRs from 21 plant species and identified Torreya grandis ANR (TgANR) as the most efficient candidate based on molecular docking and in vitro assays. Through integrated alanine scanning and computational prediction, we constructed a combinatorial mutant N15S/F106E/P182S, which exhibited a 3.88-fold higher EGC synthesis activity than the wild-type enzyme. The interaction mechanism between TgANR and the substrate delphinidin as well as the structure-activity relationship in the catalytic production of EGC were studied on the basis of protein engineering modification, thus clarifying the catalytic mechanism of TgANR. Furthermore, transient overexpression of the mutant gene in T. grandis kernels significantly increased EGC accumulation. This study not only elucidates the structure activity relationship of TgANR but also provides an efficient enzyme variant and a protein engineering framework for boosting EGC biosynthesis in plants.

Almonds (Prunus dulcis; family Rosaceae) contain 18-25% protein (dry weight). They are an important plant-based protein source in dairy alternatives and other functional foods. The hard and dense nature of almond kernels and the localization of proteins with lipid bodies in the cotyledons of almond seeds make it challenging to recover protein from the seed efficiently and preserve its function. Therefore, this review evaluates the influence of pretreatments, including blanching, grinding, and defatting, on almond protein recovery and functionality, and compares conventional and emerging technologies for almond protein. Traditional protein extraction techniques such as alkaline extraction-isoelectric precipitation (AE-IEP), aqueous extraction, and salt extraction provide moderate-to-high protein yields, but harsh processing conditions denature the proteins, decrease solubility, and cause functional properties to be lost. On the other hand, emerging protein extraction technologies (including enzyme-assisted aqueous extraction (EAE) ultrasound-assisted extraction (UAE), microwave-assisted extraction (MAE), high-pressure processing (HPP), and pulsed electric field (PEF) treatment) improve protein recovery, resulting in protein extract with superior functional properties and reduced allergenicity. However, their application in industry remain challenging. This review reveals that pretreatment approaches and conditions/parameters significantly influence protein extraction efficiency and the functional and structural properties of almonds, and that no single method is universally optimal. This review concludes that controlled enzymatic hydrolysis combined with physical pretreatment may be the best approach for producing high-value-added almond protein ingredients with specific techno-functional properties for use in plant-based beverages, hypoallergenic products, or nutraceuticals. More research is needed to develop an efficient, applicable, sustainable and eco-friendly almond protein extraction process, optimizing processing conditions to achieve high protein recovery while retaining desirable functional properties, and reduce operating costs.

Almond (Prunus dulcis) is a highly nutritious nut, rich in unsaturated fatty acids, fibre, and bioactive compounds such as tocopherols, phenolic compounds, and carotenoids. Processing methods commonly applied in the almond industry, including blanching and roasting, may modify the nutritional composition and bioactive profile of the kernels. Therefore, the aim of this study was to evaluate the effect of blanching and roasting on the nutritional composition and bioactive compound content of the 'Largueta' almond variety. Three forms were analysed: raw almonds with skin, blanched (peeled) raw almonds, and roasted almonds, with their chemical composition, lipid profile and bioactive compound content being examined. The data obtained indicated that raw almonds with skin showed higher levels of fibre (12.16 g/100 g), phenolic compounds (66.35 mg gallic acid/100 g), and &#x3b2;-carotene (65.88 &#xb5;g/100 g). Roasted almonds contained lower amounts of phenolic compounds (42.87 mg gallic acid/100 g), tocopherols (7.64 mg &#x3b1;-tocopherol/100 g and 1.99 mg &#x3b3;-tocopherol/100 g) and essential amino acids such as tryptophan (1.23 g/100 g protein) and lysine (3.22 g/100 protein). Blanching, by removing the skin, significantly reduces fibre (7.52 g/100 g) and carotenes (26.60 &#xb5;g &#x3b2;-carotene/100 g). With regard to fatty acids, the main components of nuts, oleic acid predominated in all samples (>65%), with no significant changes due to processing. Thermal treatments modify the composition of almonds. Roasting concentrates some nutrients but reduces antioxidants, while blanching mainly affects fibre content. Therefore, the consumption of raw sweet almonds with skin is recommended to preserve their nutritional and antioxidant benefits, or to subject them to moderate heat treatment.

Predicting the human burden of vector-borne diseases from limited surveillance data remains a major challenge, particularly in the presence of nonlinear transmission dynamics and delayed effects arising from vector ecology and human behavior. We develop a data-driven framework based on an extension of Sparse Identification of Nonlinear Dynamics to systems with distributed memory, enabling discovery of transmission mechanisms directly from time series data. Using severe fever with thrombocytopenia syndrome as a case study, we show that this approach can uncover key features of tick-borne disease dynamics using only human incidence and local temperature data, without imposing predefined assumptions on human case reporting. We further ascertain the robustness of the recovered incidence-temperature model by integrating it with mechanistically derived tick-host covariates, showing that the forecasting ability does not improve. This suggests that the proposed core data-driven model already delivers strong predictions. The framework also allows for systematic sensitivity analysis of memory kernels and behavioral parameters. Although the approach prioritizes predictive accuracy over mechanistic transparency, it yields sparse, interpretable integral representations suitable for epidemiological forecasting. This methodology provides a scalable strategy for forecasting vector-borne disease risk and informing public health decision-making under data limitations.

Single-lead electrocardiogram (ECG) is widely used in wearable devices for atrial fibrillation (AF) screening. Nevertheless, subtle pathological characteristics like P-waves and f-waves in practical signals are vulnerable to noise contamination. Meanwhile, the scarcity of high-quality annotated abnormal data instances leads to severe class imbalance. To mitigate these issues, we present an end-to-end framework designed for arrhythmia diagnosis using single-lead ECG signals, which integrates quality-aware data augmentation with a Peak-Enhanced attention mechanism. First, to mitigate the problem of data imbalance, a Quality-Aware Generative Adversarial Network (QA-GAN) is designed. This network integrates a signal quality evaluation module based on signal kurtosis, together with a dynamic soft-label training scheme, guiding the generator to prioritize learning high-quality morphological features, thereby synthesizing high-fidelity minority class samples. Second, to accurately capture subtle pathological features in electrocardiograms, a Peak-Enhanced Attention Convolutional Network (PEAC-Net) classification model is proposed. This model incorporates a Peak-Enhanced Attention (PE-Att) module, which employs learnable derivative convolutional kernels to precisely identify the transition points in the ECG signal. Furthermore, by integrating one-dimensional multi-scale dilated convolution (DSGC1D) with bidirectional LSTM, the model achieves effective capturing of both fine-grained local morphological features and long-range global rhythm patterns. Experimental results on the PhysioNet 2017 dataset indicate that the presented model attains an accuracy of 0.902 and a macro-F1 score of 0.880, respectively, outperforming other state-of-the-art models and also exhibiting robust data adaptability on the MIT-BIH dataset.

This study aimed to characterize the stress relaxation of wheat kernels of different hardness and varying moisture contents using generalized Maxwell models with three, five, and seven elements. Single-kernel stress relaxation tests were conducted under controlled axial compression to obtain force decay data, which were fitted using non-linear regression to estimate model constants. Positive correlations between kernel moisture content and Maxwell constants were observed, indicating greater magnitude of force decay and higher relaxation rates at higher moisture levels, while the residual force decreased. Significant differences in viscoelastic parameters were also found between hard and soft wheat varieties.

A Gaussian Process Regression (GPR) model was developed to predict the shear capacity of exterior reinforced concrete (RC) beam-column joints subjected to seismic loading. The model accounts for key parameters, including beam and column geometry, reinforcement detailing, axial column load, and concrete compressive strength. A database of 273 experimentally tested specimens was used, with emphasis on horizontal joint shear strength. Three kernel structures within the GPR framework-Primary, Rational Second-Order, and Combined kernels-were examined. Model predictions were evaluated against existing shear-strength formulations using deterministic metrics (MAE, RMSE, and R2) and probabilistic measures (NLPD and MSLL). The results show that kernel effectiveness depends on model formulation; however, the Combined kernel exhibited more stable predictions and improved uncertainty calibration for models with higher-dimensional input sets. Sensitivity analysis identified concrete compressive strength, beam depth, and joint transverse reinforcement as dominant variables, followed by column height, axial load ratio, and reinforcement configuration. Overall, the proposed framework enables consistent shear-strength prediction and quantifies the relative influence of geometric and material parameters, contributing to more informed assessment and design of RC beam-column joints.

Background/Objectives: Popcorn (Zea mays L. var. everta) is an important specialty maize type; however, the genetic variation underlying popping-related quality traits remains insufficiently characterized in breeding. Methods: In this study, 18 popcorn inbred lines were analyzed using 25 simple sequence repeat (SSR) markers distributed across all 10 maize chromosomes, and 16 lines were further evaluated for popping performance and image-based flake morphology. Results: Substantial phenotypic variation was observed among the tested lines, with expansion volume ranging from 173.33 to 343.33 mL and expandability ranging from 16.79- to 32.46-fold. Image-based analysis of 957 popped kernels revealed continuous variation in flake circularity, indicating that flake morphology represents a quantitative trait rather than a strictly discrete classification. SSR analysis detected 2 to 11 alleles per locus, with polymorphism information content values ranging from 0.05 to 0.85, indicating moderate-to-high genetic diversity among the tested lines. Principal component analysis (PCA), unweighted pair group method with arithmetic mean (UPGMA) clustering, and population structure analysis revealed clear genetic differentiation and heterogeneous genetic backgrounds within the germplasm collection. Marker-trait association analysis identified several putative SSR loci associated with expansion efficiency, flake morphology, pericarp retention, and popping dynamics. Notably, marker M18 was putatively associated with both expansion volume and expandability. Conclusions: Based on these results, a conceptual framework was proposed in which popping-related traits were organized into partially independent but interconnected functional modules. Overall, this study provides SSR-based genetic information for popcorn germplasm characterization and offers preliminary marker resources for quality-oriented popcorn breeding.

Low back pain affects hundreds of millions of people and is associated with degenerative changes in the lumbar spine. Trabecular bone microarchitecture change is a reasonable contributor to low back pain, but it remains largely invisible on routine clinical CT because typical voxel sizes (500 to 1000 &#x3bc;m) are insufficient to resolve trabeculae (about 100-200 &#x3bc;m). We present LumbarSR, a paired and registered dataset of 30 human lumbar vertebral specimens scanned with a photon-counting micro-CT (Micro-PCCT) reference at 105 &#x3bc;m isotropic resolution and with a standard clinical CT system under eight acquisition configurations formed by the factorial combination of two in-plane resolutions (195 and 586 &#x3bc;m), two slice thicknesses (500 &#x3bc;m and 1000 &#x3bc;m), and two reconstruction kernels (bone and soft tissue). Clinical CT volumes are rigidly registered to the Micro-PCCT reference using ANTs-based alignment and resampled to a common voxel grid, enabling voxel-wise evaluation and supervised learning. LumbarSR provides data in both original DICOM and registered NIfTI volumes with a consistent directory structure and specimen identifiers. We provide baseline evaluations using whole-image and masked image-quality metrics, trabecular morphometry against the Micro-PCCT reference, and super-resolution benchmarks based on interpolation and deep learning methods. LumbarSR is intended to support the development and evaluation of super-resolution methods for lumbar vertebra CT and related analyses of trabecular level structure. Because specimens were de-identified dry teaching specimens without available clinical histories or demographic metadata, LumbarSR should be interpreted as a paired imaging and benchmarking resource rather than a clinically labeled cohort.

The swift progression of three-dimensional (3D) concrete printing has opened the door to the creation of innovative materials, such as fiber-reinforced concrete, making it essential to develop accurate models for predicting their mechanical properties. Accurately estimating the compressive strength (CS) of such materials is required for optimizing mix designs and ensuring structural performance. In this context, the present study is the first to systematically investigate and compare three different support vector regression (SVR) kernels, namely, linear SVR (L-SVR), polynomial SVR (Poly-SVR), and radial basis function SVR (RBF-SVR), for predicting the compressive strength of normal, high, and ultra-high strength 3D-printed fiber-reinforced concrete (3DPFRC). A model was developed and validated using a dataset of 278 samples collected from the literature. In addition to SVR models, Artificial Neural Networks (ANN) and Gradient Boosting Machines (GBM) were developed for benchmarking purposes. Results from statistical evaluation revealed that the ANN model achieved the highest predictive accuracy overall, outperforming the L-SVR, Poly-SVR, and GBM models. Among the SVR-based models, the RBF-SVR demonstrated superior performance, showing competitive accuracy with the ANN while outperforming both the other SVR variants (L-SVR and Poly-SVR) as well as the GBM model in terms of coefficient of determination (R&#xb2;) value and error metrics. All models achieved R&#xb2; scores ranging between 0.94 and 0.86 for the training datasets and 0.93 and 0.72 in the testing datasets. The RBF-SVR model, in particular, offered an excellent balance between accuracy, training speed, and reliability, making it a strong and efficient alternative to more complex models. Additionally, SHapley Additive exPlanations (SHAP) analysis identified the water-to-cement ratio, silica fume content, and cement volume as the most influential parameters affecting compressive strength.

This study aimed to examine the associations of fine particulate matter (PM2.5) constituents and their interactions with meteorological factors with blood pressure (BP) among Chinese school-aged children, to inform targeted protection strategies. We analyzed 16,446 children from the Zhongshan Student Growth Cohort, with annual physical examinations conducted from 2006 to 2020 in Southeast China. Bayesian kernel machine regression (BKMR), weighted quantile sum (WQS), distributed lag non-linear models (DLNMs) and restricted cubic splines (RCS) were used to analyze the joint effects, exposure-response threshold, and effect modification by meteorological factors of PM2.5 constituents on BP. Overall, 20.2% were classified with high blood pressure (HBP). Each IQR increase in PM2.5 over lag0-14d was associated with a higher risk of HBP (RR&#x2009;=&#x2009;1.97, 95%CI: 1.61-2.34) among overweight children. BKMR and WQS models consistently revealed joint effects of PM2.5 constituents on BP, with Black carbon (BC) as the dominant driver. RCS models established non-linear exposure-response relationships and suggested potential threshold concentrations for key components (BC: 1.98&#xa0;&#xb5;g/m&#xb3;). Notably, both low temperature and low relative humidity (RH) significantly amplified the toxicity of PM2.5 constituents. PM2.5 constituents were adversely associated with BP in children, particularly for BC. These associations were modified by meteorological factors, with both low temperature and low RH exacerbating the adverse effects. These findings highlight the importance of considering meteorological factors in developing evidence-based air quality guidelines.

Samples of domestically distributed soybeans from Brazil produced in 2021 and 2022 were analyzed using 17 event-specific and 2 screening detection methods for genetically modified (GM) soybeans. We found that the detected varieties of GM soybeans from Brazil were significantly different from those observed in our previous study on GM soybeans imported from the United States and Canada within the same years [Soga et al., GM Crops & Food, 16, 116-125 (2025)]. After analyzing five different lots for each year, RRS, MON89788, and MON87701 were detected in all lots, whereas MON87708 and MON87751 were detected in four lots. RRS, MON89788, MON87701, MON87708 and MON87751 are a glyphosate tolerant, a glyphosate tolerant, an insect resistant, a glyphosate and dicamba tolerant, and an insect resistant GM soybean, respectively. Single kernel-based analyses revealed that stacked varieties, including MON89788 and MON87701, comprised a large proportion of GM soybeans from Brazil.

The deep integration of breakthrough technological innovation and healthcare has given rise to new quality productive forces in healthcare, profoundly reshaping the spatial logic of medical resource allocation. However, existing studies lack an evaluation framework tailored to this emerging concept, and its spatiotemporal evolution and regional coordination remain insufficiently explored, hindering targeted policy-making for balanced healthcare development. We measured the level of new quality productive forces in healthcare across 31 Chinese provinces during 2010-2023 using the CRITIC-entropy combined weighting method and TOPSIS method. Dagum Gini coefficient and kernel density analysis were applied to explore its spatial differentiation and temporal evolution characteristics. &#x3c3; convergence, absolute &#x3b2; and conditional &#x3b2; convergence models were adopted to clarify its convergence pattern. China's new quality productive forces in healthcare showed an "initial decline then sustained growth" trend, peaking in 2023. New quality medical laborers exhibited the strongest performance, while new quality medical objects of labor were the weakest. Spatially, the Eastern Region had the highest level in 2023 and the Western Region recorded the lowest level; regional gaps narrowed steadily. Inter-group disparities dominated overall disparities during 2011-2019, while transvariation was the main source in 2010 and post-2020. Absolute &#x3b2; and conditional &#x3b2; convergence were observed, but no &#x3c3; convergence existed. China's new quality productive forces in healthcare have shown dynamic growth, yet they continue to face structural constraints characterized by persistently underdeveloped new quality medical objects of labor and narrowing yet non-negligible regional disparities. Our findings highlight the need for an integrated policy approach that involves implementing differentiated strategies aligned with the heterogeneous development levels and convergence characteristics of healthcare new quality productive forces across regions, while enhancing patient digital engagement and expanding the accessibility of high-quality online healthcare services.

Neural tube defects (NTDs) exhibit a multifaceted etiology. Limited research has assessed the association between exposure to metallic and non-metallic elements and the incidence of NTDs at the elementomic level. Our study included 40 women with NTD-affected pregnancies and 119 controls in northern China's Shanxi province. Inductively coupled plasma mass spectrometry was employed to quantify 45 elements in umbilical cord serum samples collected from all participants. Thirteen elements were excluded from subsequent analysis due to a detection rate below 60%. We employed three machine learning selection models, Boruta, Least Absolute Shrinkage and Selection Operator, and Extreme Gradient Boosting, to jointly identify the key exposure elements as zinc (Zn), sodium (Na), molybdenum (Mo), cerium (Ce), barium (Ba) and titanium (Ti). Bayesian kernel machine regression assessed both single and combined effects of these elements on NTD risk, with individual element effects confirmed via logistic regression. Our results showed a non-significant trend toward higher NTD risk with rising concentrations of the mixture of these six elements. When Ce concentration increased from the 25th to the 75th percentile, a statistically significant elevated risk of NTDs was observed, after adjusting for the concentrations of the other five metals at the 25th, 50th and 75th percentiles. Logistic regression analysis further identified a significant association between Ce exposure and NTD risk, with an odds ratio (95% CI) of 1.96 (1.10-3.49). No interactions among the elements were identified. In conclusion, our results indicate that higher Ce concentrations are significantly associated with NTD risk.

This study systematically investigates the hot deformation behavior of TC18 alloy under the conditions of deformation temperatures of 720-840 &#xb0;C and strain rates of 0.001-1 s-1. Based on the stress-strain data obtained under the aforementioned process parameters, a support vector regression (SVR) model was established and further optimized by using a Stacking algorithm to enhance predictive accuracy. Although SVR and Stacking techniques have been applied previously in material constitutive modeling, this paper presents a systematic optimization framework specifically for TC18, integrating comprehensive experimental data, kernel selection, hyperparameter tuning, and Stacking-based model fusion. The polynomial kernel function was identified as optimal, and hyperparameters were tuned via grid search combined with five-fold cross-validation, which is determined as {C = 1000, coef0 = 1, d = 5, &#x3b5; = 1, &#x3b3; = 1}. The Stacking-SVR model exhibits significantly improved fitting and generalization performance compared to Poly-SVR, Arrhenius, XGBoost and MLP, with RMSE, MAPE, and R2 metrics of 2.7882, 0.0110, and 0.9973 on the training set, and 2.7956, 0.0169, and 0.9982 on the test set, respectively. Additionally, the proportion of samples with relative errors within 5% reaches 98.7% for the training set and 94.83% for the test set. These results indicate that the proposed framework not only possesses extremely high predictive accuracy, but also ensures strong generalization ability and interpretability in practical applications.

Prenatal exposure to individual metals has been studied, but effects of two-timepoint metal mixtures on offspring blood pressure (BP) and their mechanisms remain unclear. The prospective birth cohort study (n=541) investigated the effects of two-timepoint prenatal metal mixture exposure on offspring blood pressure (BP) and its underlying metabolic mechanisms. Measuring eight toxic metals in maternal blood during early to mid-pregnancy and umbilical cord blood at delivery, combined with neonatal metabolomics analysis, the study revealed a positive association between maternal metal mixture exposure and offspring BP. Using Bayesian kernel machine regression (BKMR) and weighted quantile sum (WQS) models, arsenic (As) was identified as the most significant contributor (PIP>0.8 in BKMR, weights>0.5 in WQS). Both maternal blood As and cord blood As showed significant positive associations with offspring systolic BP (SBP) [Pregnancy: &#x3b2;=4.41, 95% CI 2.39-6.44, P<0.001; Cord: &#x3b2;=2.97, 95% CI 1.17-4.78, P=0.013] and mean arterial pressure (MAP) [Pregnancy: &#x3b2;=2.67, 95% CI 1.13-4.21, P<0.001; Cord: &#x3b2;=2.50, 95% CI 1.06-3.94, P<0.001]. Metabolic pathway analysis indicated that prenatal arsenic exposure was associated with alterations in carnitine homeostasis (including acetylcarnitine [C2], suberylcarnitine [C8DC], decanoylcarnitine [C10], and decadienoylcarnitine [C12DC]), while showing little association with amino acid metabolism. These carnitine-related metabolic changes were consistent with a potential mediating role in the association between prenatal arsenic exposure and offspring blood pressure. Early to mid-pregnancy exposure to metal mixtures, particularly arsenic, shows a stronger association with offspring BP than later-gestation exposure measures, and this association is linked to disruptions in carnitine metabolism.