Search Results (845)

Accurate interpretation of missense variants in cancer-associated genes remains a critical challenge in precision oncology, as most sequence-based predictors lack mechanistic explanations. Receptor tyrosine kinases like FGFR2 exemplify this problem: their function depends on conformational dynamics, yet most variants remain classified as variants of uncertain significance (VUS). In this paper we present DyVarMap, an interpretable structural-learning framework that integrates AlphaFold2-based ensemble generation with physics-driven refinement, manifold learning, and supervised classification using five biophysically motivated geometric features. Applied to FGFR2, the framework generates diverse conformational ensembles, identifies metastable states through nonlinear dimensionality reduction, and classifies pathogenicity while providing mechanistic attributions via SHAP analysis. External validation on ten kinase-domain variants yields an AUROC of 0.77 with superior calibration (Brier score = 0.108) compared to PolyPhen-2 (0.125) and AlphaMissense (0.132). Feature importance analysis consistently identifies K659–E565 salt-bridge distance and DFG motif dihedral angles as top predictors, directly linking predictions to known activation mechanisms. Case studies of borderline variants (A628T, E608K, L618F) demonstrate the framework’s ability to provide structurally coherent mechanistic explanations. DyVarMap bridges the gap between static structure prediction and dynamics-aware functional assessment, generating testable hypotheses for experimental validation and demonstrating the value of incorporating conformational dynamics into variant effect prediction for precision oncology. Full article

Catalytic pyrolysis (CP) of biomass is a promising method for producing sustainable hydrogen because lignocellulosic biomass is widely available, renewable, and approximately carbon-neutral. CP of biomass is influenced by complex, interdependent process parameters, making optimization challenging and time-consuming using traditional methods. This study investigated a two-stage machine learning (ML) framework fortified with Bayesian optimization to enhance hydrogen production from CP. The ML models were used to classify and predict hydrogen yield using a dataset of 306 points with 14 input features. The classification stage identified conditions favorable for good hydrogen yield, while the regression model (second stage) quantitatively predicted hydrogen yield. The random forest classifier and regressor demonstrated superior capabilities, achieving R² scores of 1.0 and 0.8, respectively. The model demonstrated strong agreement with experimental data and effectively captured the key factors driving hydrogen production. Shapley Additive exPlanation (SHAP) identified temperature and catalyst properties (nickel loading) as the most influential parameters. The inverse analysis framework validated the model’s ability to determine optimal conditions for predicting targeted hydrogen yields by comparing it to experimental data reported in the literature. This AI-driven approach provides a scalable and data-efficient tool for optimizing processes in sustainable hydrogen production. Full article

Genomic selection (GS) estimates the GEBV from genome-wide markers to reduce generation intervals and optimize germplasm selection, which is particularly advantageous for high-cost or late-expressed traits. While models like GBLUP are popular, they assume a polygenic architecture. In contrast, the Bayesian alphabet and machine learning (ML) can accommodate other types of genetic architectures. Given that no single model is universally optimal, stacking ensembles, which train a meta-model using predictions from diverse base learners, emerge as a compelling solution. However, the application of stacking in GS often overlooks non-additive effects. This study evaluated different stacking configurations for genomic prediction across 10 simulated traits, covering additive, dominance, and epistatic genetic architectures. A 5-fold cross-validation scheme was used to assess predictive ability and other evaluation metrics. The stacking approach demonstrated superior predictive ability in all scenarios. Gains were especially pronounced in complex architectures (100 QTLs, h² = 0.3), reaching an 83% increment over the best individual model (BayesA with dominance), and also in oligogenic scenarios with epistasis (10 QTLs, h² = 0.6), with a 27.59% gain. The success of stacking was attributed to two key strategies: base learner selection and the use of robust meta-learners (such as principal component or penalized regression) that effectively handled multicollinearity. Full article

Background: The persistently low 5-year survival rate for pancreatic cancer (PC) underscores the critical need for early detection. However, population-wide screening remains impractical. Artificial Intelligence (AI) models using electronic health record (EHR) data offer a promising avenue for pre-symptomatic risk stratification. Objective: To systematically review and meta-analyze the performance of AI models for PC prediction based exclusively on structured EHR data. Methods: We systematically searched PubMed, MedRxiv, BioRxiv, and Google Scholar (2010–2025). Inclusion criteria encompassed studies using EHR-derived data (excluding imaging/genomics), applying AI for PC prediction, reporting AUC, and including a non-cancer cohort. Two reviewers independently extracted data. Random-effects meta-analysis was performed for AUC, sensitivity (Se), and specificity (Sp) using R software version 4.5.1. Heterogeneity was assessed using I² statistics and publication bias was evaluated. Results: Of 946 screened records, 19 studies met the inclusion criteria. The pooled AUC across all models was 0.785 (95% CI: 0.759–0.810), indicating good overall discriminatory ability. Neural Network (NN) models demonstrated a statistically significantly higher pooled AUC (0.826) compared to Logistic Regression (LogReg, 0.799), Random Forests (RF, 0.762), and XGBoost (XGB, 0.779) (all p < 0.001). In analyses with sufficient data, models like Light Gradient Boosting (LGB) showed superior Se and Sp (99% and 98.7%, respectively) compared to NNs and LogReg, though based on limited studies. Meta-analysis of Se and Sp revealed extreme heterogeneity (I² ≥ 99.9%), and the positive predictive values (PPVs) reported across studies were consistently low (often < 1%), reflecting the challenge of screening a low-prevalence disease. Conclusions: AI models using EHR data show significant promise for early PC detection, with NNs achieving the highest pooled AUC. However, high heterogeneity and typically low PPV highlight the need for standardized methodologies and a targeted risk-stratification approach rather than general population screening. Future prospective validation and integration into clinical decision-support systems are essential. Full article

Facial Beauty Prediction (FBP) aims to develop a machine that can automatically predict facial attractiveness. Recent advances demonstrate that deep learning models have achieved promising results in FBP tasks. However, conventional deep learning models lack the efficient ability to generalize to unseen attribute domain data, as attributes cause distribution discrepancy (asymmetry) among face data. To address this issue, we propose a simple yet effective method called MixAttr, a lightweight plug-and-play module that mixes two randomly selected feature statistics with different attributes to form a newly attributed feature. In this way, the feature space is implicitly enriched by increasing the diversity of features and mitigating the model shift caused by a single attribute, which benefits the decoupling of attributes and facial beauty prediction. Extensive experiments conducted to evaluate the properties and effectiveness of our method show that MixAttr can be flexibly inserted into existing network architectures to achieve state-of-the-art performance on different FBP benchmarks (e.g., a Pearson correlation of 0.9307 on SCUT-FBP5500). This feature mixing implicitly enriches the representation space, which is key to mitigating attribute-induced asymmetry and improving generalization. Additionally, we have also extended our method to the task of facial age estimation, demonstrating through superior experimental results that our method can also be applied to other attribute prediction tasks. We propose that distribution discrepancy in FBP can be viewed as a form of asymmetry in the feature space across different demographic groups. MixAttr mitigates this asymmetry by implicitly enriching the feature space and encouraging a more symmetric and attribute-invariant feature representation. By disentangling task-irrelevant attributes from task-oriented features, our method can improve both the accuracy and generalizability of deep models on tasks involving facial attribute prediction. Full article

An accurate assessment of the state of health (SOH) of lithium-ion batteries (LIBs) is crucial for ensuring the safety and reliability of energy storage systems and electric vehicles. However, existing methods face challenges: physics-based models are computationally complex, traditional data-driven methods rely heavily on manual feature engineering, and single models lack the ability to capture both local and global degradation patterns. To address these issues, this paper proposes a novel hybrid LSTM–Transformer model for LIB SOH estimation using actual measurement data. The model integrates Long Short-Term Memory (LSTM) networks to capture local temporal dependencies with the Trans-former architecture to model global degradation trends through self-attention mechanisms. Experimental validation was conducted using eight 18650 Nickel Cobalt Manganese (NCM) LIBs subjected to 750 charge–discharge cycles under room temperature conditions. Sixteen statistical features were extracted from voltage and current data during constant current–constant voltage (CC-CV) phases, with feature selection based on the Pearson correlation coefficient and maximum information coefficient analysis. The proposed LSTM–Transformer model demonstrated superior performance compared to the standalone LSTM and Transformer models, achieving a mean absolute error (MAE) as low as 0.001775, root mean square error (RMSE) of 0.002147, and mean absolute percentage error (MAPE) of 0.196% for individual batteries. Core features including cumulative charge (CC Q), charging time, and voltage slope during the constant current phase showed a strong correlation with the SOH (absolute PCC > 0.8). The hybrid model exhibited excellent generalization across different battery cells with consistent error distributions and nearly overlapping prediction curves with actual SOH trajectories. The symmetrical LSTM–Transformer hybrid architecture provides an accurate, robust, and generalizable solution for LIB SOH assessment, effectively overcoming the limitations of traditional methods while offering potential for real-time battery management system applications. This approach enables health feature learning without manual feature engineering, representing an advancement in data-driven battery health monitoring. Full article

Chronic obstructive pulmonary disease (COPD) is a major health burden worldwide and in Taiwan, ranking as the third leading cause of death globally, and its prevalence in Taiwan continues to rise. Readmission within 14 days is a key indicator of disease instability and care efficiency, driven jointly by patient-level physiological vulnerability (such as reduced lung function and multiple comorbidities) and healthcare system-level deficiencies in transitional care. To mitigate the growing burden and improve quality of care, it is urgently necessary to develop an AI-based prediction model for 14-day readmission. Such a model could enable early identification of high-risk patients and trigger multidisciplinary interventions, such as pulmonary rehabilitation and remote monitoring, to effectively reduce avoidable early readmissions. However, medical data are commonly characterized by severe class imbalance, which limits the ability of conventional machine learning methods to identify minority-class cases. In this study, we used real-world clinical data from multiple hospitals in Kaohsiung City to construct a prediction framework that integrates data generation and ensemble learning to forecast readmission risk among patients with chronic obstructive pulmonary disease (COPD). CTGAN and kernel density estimation (KDE) were employed to augment the minority class, and the impact of these two generation approaches on model performance was compared across different augmentation ratios. We adopted a stacking architecture composed of six base models as the core framework and conducted systematic comparisons against the baseline models XGBoost, AdaBoost, Random Forest, and LightGBM across multiple recall thresholds, different feature configurations, and alternative data generation strategies. Overall, the results show that, under high-recall targets, KDE combined with stacking achieves the most stable and superior overall performance relative to the baseline models. We further performed ablation experiments by sequentially removing each base model to evaluate and analyze its contribution. The results indicate that removing KNN yields the greatest negative impact on the stacking classifier, particularly under high-recall settings where the declines in precision and F1-score are most pronounced, suggesting that KNN is most sensitive to the distributional changes introduced by KDE-generated data. This configuration simultaneously improves precision, F1-score, and specificity, and is therefore adopted as the final recommended model setting in this study. Full article

Meteorological data from automatic weather stations (AWS) in tea plantations is critical for agricultural management, but is often compromised by sensor errors and physical implausibilities that traditional quality control (QC) methods fail to detect. This study proposes a novel, meteorologically informed QC framework that mines association rules from long-term ERA5 reanalysis data (2012–2023) using the Apriori algorithm to establish a knowledge base of normal multivariate atmospheric patterns. A comprehensive feature engineering process generated temporal, physical, and statistical features, which were discretized using meteorological thresholds. The mined rules were filtered, prioritized, and integrated with hard physical constraints. The system employs a fuzzy logic mechanism for violation assessment and a weighted anomaly scoring system for classification. When validated on a synthetic dataset with injected anomalies, the method significantly outperformed traditional QC techniques, achieving an F1-score of 0.878 and demonstrating a superior ability to identify complex physical inconsistencies. Application to an independent historical dataset from a Zhenjiang tea plantation (2008–2016) successfully identified 14.6% anomalous records, confirming the temporal transferability and robustness of the approach. This framework provides an accurate, interpretable, and scalable solution for enhancing the quality of meteorological data, with direct implications for improving the reliability of frost prediction and pest management in precision agriculture. Full article

A novel stochastic finite element model updating method based on multi-point sensitivities is proposed to improve the reproduction and prediction ability of finite element models for experimental data. Drawing upon the theory of small perturbations, this approach employs the sensitivity matrix in conjunction with the probability distribution of responses evaluated at multiple parameter points to determine the probability density associated with each parameter point and to estimate the statistical properties of the parameters. To achieve this objective, principal component analysis is employed to unify the dimensionality of the parameters and the responses; the least squares method was used to estimate the characteristics of the parameters. The reliability and validity of this method were confirmed through experimentation with a 3-degree-of-freedom spring-mass system and an aerospace thermal insulation structure. A comparison of this method with classical methods reveals significant advantages in terms of robustness across varying computational scales. Notably, it attains superior accuracy with smaller sample sizes while maintaining precision comparable to conventional methods with large samples. Consequently, this characteristic confers upon the method a distinct advantage in scenarios where the costs of finite element computation are prohibitively high. Full article

The accurate prediction of human movement direction plays a crucial role in fields such as rehabilitation monitoring, sports science, and intelligent military systems. Based on plantar pressure and inertial sensor data, this study developed a hybrid deep learning model integrating a Convolutional Neural Network (CNN) and a Bidirectional Long Short-Term Memory (BiLSTM) network to enable joint spatiotemporal feature learning. Systematic comparative experiments involving four distinct deep learning models—CNN, BiLSTM, CNN-LSTM, and CNN-BiLSTM—were conducted to evaluate their convergence performance and prediction accuracy comprehensively. Results show that the CNN-BiLSTM model outperforms the other three models, achieving the lowest RMSE (0.26) and MAE (0.14) on the test set, with an R² of 0.86, which indicates superior fitting accuracy and generalization ability. The superior performance of the CNN-BiLSTM model is attributed to its ability to effectively capture local spatial features via CNN and model bidirectional temporal dependencies via BiLSTM, thus demonstrating strong adaptability for complex motion scenarios. This work focuses on the optimization and comparison of deep learning algorithms for spatiotemporal feature extraction, providing a reliable framework for real-time human motion prediction and offering potential applications in intelligent gait analysis, wearable monitoring, and adaptive human–machine interaction. Full article

The contemporary agricultural market is profoundly volatile, where agricultural prices are based on a complex supply chain, climatic irregularity or unscheduled market demand. Prices of crops need to be predicted in a reliable and timely manner for farmers, policy-makers and other stakeholders to take evidence-based decisions ultimately for the benefit towards sustainable agriculture and economic sustainability. Objective: The objective of this study is to develop and evaluate a comprehensive machine learning model for predicting agricultural prices incorporating logistic, economic and environmental considerations. It is the desire to make agriculture more profitable by building simple and accurate forecasting models. Methods: An assorted dataset was collected, which covers major factors to constitute the dataset of temperature, rainfall, fertiliser use, pest and disease attack level, cost of transportation, market demand-supply ratio and regional competitiveness. The data was subjected to pre-processing and feature extraction for quality control/quality assurance. Several machine learning models (Linear Regression, Support Vector Machines, AdaBoost, Random Forest, and XGBoost) were trained and evaluated using performance metrics such as R2 score, Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). Results: Out of the model approaches that were analysed, predictive performance was superior for XGBoost (with an R2 Score of 0.94, RMSE of 12.8 and MAE of 8.6). To generate accurate predictions, the ability to account for complex non-linear relationships between market and environmental information was necessary. Conclusions: The forecast model of the XGBoost-based prediction system is reliable, of low complexity and widely applicable to large-scale real-time forecasting of agricultural monitoring. The model substantially reduces the uncertainty of price forecasting, and does so by including multivariate environmental and economic aspects that permit more profitable management practices in a schedule for future sustainable agriculture. Full article

As the core technology of clean and efficient utilization of coal, coal gasification technology plays an important role in reducing environmental pollution, improving coal utilization, and achieving sustainable energy development. In order to ensure the safe, stable, and long-term operation of coal gasification plant, aiming to address the strong subjectivity of dynamic Bayesian network (DBN) prior data in dynamic risk assessment, this study takes the coal slurry preparation system—the main piece of equipment in the initial stage of the coal gasification process—as the research object and uses a long short-term memory (LSTM) model combined with a back propagation (BP) neural network model to optimize DBN prior data. To further validate the superiority of the model, a gated recurrent unit (GRU) model was introduced for comparative verification. The mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination are used to evaluate the generalization ability of the LSTM model. The results show that the LSTM model’s predictions are more accurate and stable. Bidirectional inference is performed on the DBN of the optimized coal slurry preparation system to achieve dynamic reliability analysis. Thanks to the forward reasoning of DBN in the coal slurry preparation system, quantitative analysis of the system’s reliability effects is conducted to clearly demonstrate the trend of system reliability over time, providing data support for stable operation and subsequent upgrades. By conducting reverse reasoning, key events and weak links before and after system optimization can be identified, and targeted improvement measures can be proposed accordingly. Full article

Forest fire susceptibility prediction is essential for effective management, yet considerable uncertainty persists under future climate change, especially in climate-sensitive plateau regions. This study integrated MODIS fire data with climatic, topographic, vegetation, and anthropogenic variables to construct an Extreme Gradient Boosting (XGBoost) model for the Yunnan Plateau, a region highly prone to forest fires. Compared with Support Vector Machine and Random Forest models, XGBoost showed superior ability to capture nonlinear relationships and delivered the best performance, achieving an AUC of 0.907 and an overall accuracy of 0.831. The trained model was applied to climate projections under SSP1-2.6, SSP2-4.5, and SSP5-8.5 to assess future fire susceptibility. Results indicated that high-susceptibility periods primarily occur in winter and spring, driven by minimum temperature, average temperature, and precipitation. High-susceptibility areas are concentrated in dry-hot valleys and mountain basins with elevated temperatures and dense human activity. Under future climate scenarios, both the probability and spatial extent of forest fires are projected to increase, with a marked expansion after 2050, especially under SSP5-8.5. Although the XGBoost model demonstrates strong generalizability for plateau regions, uncertainties remain due to static vegetation, coarse anthropogenic data, and differences among climate models. Full article

This work proposes an analysis of the capability of three deep learning models—the feedforward neural network (FFNN), long short-term memory (LSTM) network, and physics-informed neural network (PINN)—to identify the parameters of a flexible fixed-wing aircraft using in-flight data. These neural networks, composed of multiple hidden layers, are evaluated for their ability to perform system identification and to capture the nonlinear and dynamic behavior of the aircraft. The FNN and LSTM models are compared to assess the impact of temporal dependency learning on parameter estimation, while the PINN integrates prior knowledge of the system’s governing of ordinary differential equations (ODEs) to enhance physical consistency in the identification process. The objective is to exploit the generalization capability of neural network-based models while preserving the accurate estimation of the physical parameters that characterize the analyzed system. The neural networks are evaluated for their ability to perform system identification and capture the nonlinear behavior of the aircraft. The results show that the FFNN achieved the best overall performance, with average Theil’s inequality coefficient (TIC) values of 0.162 during training and 0.386 during testing, efficiently modeling the input-output relationships but tending to fit high-frequency measurement noise. The LSTM network demonstrated superior noise robustness due to its temporal filtering capability, producing smoother predictions with average TIC values of 0.398 (training) and 0.408 (testing), albeit with some amplitude underestimation. The PINN, while successfully integrating physical constraints through pretraining with target aerodynamic derivatives, showed more complex convergence, with average TIC values of 0.243 (training) and 0.475 (testing), and its estimated aerodynamic coefficients differed significantly from the conventional values. All three architectures effectively captured the coupled rigid-body and flexible dynamics when trained with distributed wing sensor data, demonstrating that neural network-based approaches can model aeroelastic phenomena without requiring explicit high-fidelity flexible-body models. This study provides a comparative framework for selecting appropriate neural network architectures based on the specific requirements of aircraft system identification tasks. Full article

Objective: To develop and validate an interpretable deep learning model based on the TabNet architecture for predicting doxorubicin-induced cardiotoxicity (DIC) in patients with breast cancer through integration of multidimensional clinical data. Methods: This retrospective study included 2034 patients who received doxorubicin-based chemotherapy at The Fourth Affiliated Hospital of Harbin Medical University between January 2021 and December 2023. Clinical, biochemical, electrocardiographic, and echocardiographic parameters were incorporated into six predictive algorithms: logistic regression, decision tree, random forest, gradient boosting machine, XGBoost, and TabNet. Model discrimination, calibration, and clinical utility were assessed using AUC, C-index, calibration plots, and decision curve analysis. Model interpretability was evaluated through attention-based feature importance and SHAP analysis. Results: TabNet achieved the best overall predictive performance, with an AUC of 0.86 and a C-index of 0.80 in the validation cohort, demonstrating superior discrimination, calibration, and generalization compared with all baseline models. Decision curve analysis confirmed its higher net clinical benefit across threshold probabilities. The model identified eight dominant predictors—cumulative anthracycline dose, LVEF, QTc interval, lactate dehydrogenase, creatinine, glucose, hypertension, and platelet count—that collectively reflected myocardial contractility, electrophysiological stability, and systemic metabolic stress. Correlation and clustering analyses revealed that high-risk patients exhibited concurrent QTc prolongation, metabolic disturbance, and LVEF decline, defining a distinct cardiometabolic injury phenotype. These findings highlight TabNet’s ability to uncover complex feature interactions while maintaining transparent and clinically interpretable outputs. Conclusions: The TabNet-based multidimensional model provides an accurate, stable, and interpretable tool for individualized prediction of doxorubicin-induced cardiotoxicity, supporting early intervention and precision management in breast cancer patients receiving anthracycline therapy. Full article