Search Results (9,569)

Background/Objectives: The programmed death-1/programmed death-ligand 1 (PD-1/PD-L1) immune checkpoint is a key therapeutic target in cancer immunotherapy, but small-molecule inhibition remains challenging due to its shallow and dynamic interaction interface. This study aimed to develop an artificial intelligence (AI)-guided workflow to identify novel small-molecule inhibitors targeting the PD-L1 dimer interface. Methods: A combined computational and experimental approach was established. A support vector regression-genetic algorithm (SVR-GA) model was trained on a dataset of 1385 known PD-L1 inhibitors to predict activity and guide molecular generation. From 470 AI-generated candidates, docking and molecular dynamics (MD) simulations were used for virtual screening. Selected compounds were synthesized and evaluated for PD-1/PD-L1 binding disruption using homogeneous time-resolved fluorescence (HTRF) assays. Cytotoxicity was tested in MDA-MB-231 and 4T1 cell monocultures, and in vivo efficacy was assessed in an immunocompetent 4T1 tumor model. Results: Two hits, PD-L1-Ser and PD-L1-Ser-OEt, were identified. Both disrupted PD-1/PD-L1 binding in HTRF assays, with PD-L1-Ser-OEt showing higher potency (IC₅₀ = 0.2068 μM). Both compounds exhibited limited direct cytotoxicity in cancer cell monocultures, suggesting an immune-mediated mechanism. In the 4T1 syngeneic mouse model, both inhibitors suppressed tumor growth without causing body weight loss. PD-L1-Ser-OEt demonstrated superior antitumor efficacy and elevated serum levels of IFN-γ and IL-4. Conclusions: This AI-guided workflow combining machine-learning-based molecular generation with structure validation is feasible for discovering PD-L1 dimer-interface inhibitors. PD-L1-Ser-OEt represents a promising lead compound for further development as an immune checkpoint inhibitor. Full article

This study proposes Uncertainty-Aware Staged Trust-Adaptive Residual Physics-Informed Machine Learning (U-STAR-PIML), a recursive one-step dynamics model for fixed-wing unmanned aerial vehicles, evaluated with the JSBSim C172x as a surrogate simulation benchmark. It combines a learnable compact six-degree-of-freedom prior, history-dependent temporal residual correction, trust-aware state-dependent residual gating, hard kinematics, a heteroscedastic one-step uncertainty head, and staged optimization. The protocol separates one-step accuracy, recursive rollout, predictive-interval behavior, and physically distinct distribution shifts. Across five training seeds, the full U-STAR-PIML model (E5) achieves a mean one-step root-mean-square error (RMSE) of $0.005515\pm 0.000009$ and the lowest mean rollout-position RMSE of $18.61\pm 2.15$ m; the data-driven baseline has the lowest 20 s all-state RMSE of $2.923\pm 0.623$. Recursive rankings remain seed-sensitive, without a universal winner. For the representative E5 cross-condition evaluation, the exact wind vector used in the JSBSim simulation is supplied to the model at every prediction step, i.e., perfect wind information is assumed. Under this assumption, wind out-of-distribution (OOD) conditions cause the largest degradation, with rollout-position RMSE reaching 72.76 m; wind-estimation error is not evaluated. An external zero-shot evaluation on 10 independent IDF-DS Ranger 2400 real-flight logs reduces pooled one-step all-state RMSE from 0.12374 for Persistence to 0.03379, although improvements are not uniform across dynamic state groups. The uncertainty head yields 95% empirical coverage of 96.68–100%, with conservative over-coverage under most conditions. These results support simulation-based prediction and an initial cross-airframe transfer diagnostic but do not establish same-airframe sim-to-real transfer, recursive real-flight stability, or operational validity. Full article

As the crisis of antibiotic resistance escalates, phage therapy has regained attention as an alternative strategy. Artificial intelligence (AI) technologies offer new avenues to overcome the bottlenecks inherent in traditional bacteriophage research. This review summarizes the multi-dimensional innovative applications of machine learning, deep learning, and large biological models in phage studies. In the fields of phage recognition and genomics, support vector machines (SVMs), convolutional neural networks (CNNs), and pre-trained protein language models can all achieve recognition accuracy rates of over 90%. Furthermore, tools such as DeepHost and VirSorter2 can efficiently identify phage sequences, annotate functional genes, and predict hosts at the species or strain levels. For clinical translation, AI integrates patient characteristics, bacterial phenotypes, and phage profiles to customize cocktail regimens for individualized phage therapy. Graph neural network-based models like DeepPBI-KG integrate multi-omics knowledge graphs to precisely predict phage-host interactions (PHIs), whereas agent-based simulation and defense protein predictors forecast phage resistance evolution. Additionally, generative AI can support the de novo design of functional phage genomes and mine massive unannotated virome dark matter. Nevertheless, this cross-disciplinary field faces significant constraints, including uneven and biased sequencing datasets, insufficient model interpretability, and dual-use biosafety ethical risks accompanied by unclear algorithm accountability and incomplete global supervision systems. Future research should optimize standardized multimodal databases, develop explainable AI algorithms, and establish cross-disciplinary ethical governance frameworks to facilitate closed-loop verification between computational prediction and wet-lab experiments. In conclusion, the deep integration of AI and phage biology provides revolutionary strategies to tackle multidrug-resistant infections and advances the clinical transformation of phage precision medicine. Full article

Accurate identification of runway base-course parameters is an important basis for structural condition assessment, yet separating base-course elastic modulus, thickness, and local damage from runway responses under moving aircraft loads remains largely unexplored. This study proposes a machine learning inversion method based on multi-domain vibration features. Dynamic responses are generated with a 2.5-dimensional finite-element–boundary-element model, from which peak strain, dominant frequency, low-frequency energy ratio, strain attenuation gradient, and time–frequency entropy are extracted; random forest (RF) and support vector machine (SVM) models then invert three classes of base-course parameters. The extracted features exhibited distinct sensitivities to base-course modulus, thickness, and local damage. RF and SVM showed different advantages depending on the evaluation metric, while SVM performed better in the unseen interpolation cases. The modulus-identification performance of both models remained comparatively stable after noise was added. The results indicate that multi-domain vibration features provide complementary information for the numerical inversion of runway base-course parameters, thereby establishing a methodological basis for their rapid assessment. Full article

The timberline region of the Western Himalaya is a crucial ecological transition area highly sensitive to climate changes, which significantly influence vegetation patterns, soil formation, and carbon dynamics. This study aimed to investigate the spatial and altitudinal changes in soil organic carbon (SOC) across different topographic orientations and to evaluate machine-learning models for spatial SOC prediction in the timberline ecotone (2100–3300 m) of the Kedarnath Wildlife Sanctuary. Through systematic random sampling across 100 m elevation bands, composite soil samples were collected from three aspects (North-East, South-West, and North-West) at two depths (0–15 cm and 15–30 cm) and analyzed alongside topographic, spectral, and climatic covariates. Results indicated that SOC trends varied significantly by aspect; the North-East aspect exhibited a considerable increase in SOC with elevation, while the North-West and South-West sides responded differently. Furthermore, Digital Soil Mapping using the Random Forest (RF) model outperformed Support Vector Machine and XGBoost, explaining 62% of surface and 74% of subsurface SOC variability. In conclusion, aspect-induced microclimatic gradients play a major role in controlling soil characteristics near the Himalayan timberline, and machine-learning models like RF can successfully capture these complex spatial patterns. Consequently, it is recommended that the established baseline SOC maps be utilized as reference points for future climate-change monitoring, carbon accounting, and directing conservation strategies in high-altitude forests. Full article

This study addresses a regime-conditioned question in electricity theft detection: when only monthly billing series and inspection-confirmed labels are available, which supervised model families recover irregular consumption without relying on the temporal resolution of advanced metering infrastructure (AMI)? The working sample comprises 4000 utility customers and 864 confirmed theft cases, each represented by 53 monthly kWh values from January 2021 to May 2025. After majority-class undersampling that retains all theft observations and construction of a balanced 1:1 learning set, eight classifiers are compared under an 80/20 stratified split: K-Nearest Neighbors, Decision Tree, Support Vector Machine, Random Forest, two dense multilayer perceptrons, Long Short-Term Memory, and a one-dimensional Convolutional Neural Network. Performance is assessed through threshold-optimized accuracy together with precision, recall, F1-score, the area under the receiver operating characteristic curve (AUC), and confusion matrices. On the hold-out test set, Random Forest and the compact dense network both reach an accuracy of 0.685; Random Forest attains the highest AUC (0.748) and F1-score (0.677). Even so, these models miss about one-third of the hold-out theft accounts (59 and 65 false negatives out of 173). Sequential deep models underperform on this short monthly regime. The results support ensembles and compact dense networks for monthly theft screening and indicate that AMI-oriented sequential gains do not transfer automatically to 53-point billing vectors under the present protocol. Full article

Loess mountainous regions in northern China suffer frequent landslides, collapses and debris flows controlled by complex geological settings, seasonal rainstorms, freeze–thaw cycles and large-scale human engineering activities. Multi-geohazard susceptibility evaluation can provide fundamental data support for regional disaster prevention and territorial planning. Taking Shanxi Province, a typical loess-mountain transition zone, as the study area, this paper establishes an evaluation framework for landslides, collapses and debris flows. Ten conditioning factors are selected, including lithology, terrain parameters, distance to faults, distance to rivers, NDVI and RX1day (annual maximum 1-day precipitation). A 30 m grid unit is adopted as the basic evaluation unit. A total of 2598 verified geohazard points are taken as positive samples. Negative samples with equal quantity are extracted from low and very low susceptibility areas of the preliminary zoning map generated by the Frequency Ratio (FR) method, with an 800 m minimum separation distance between sampling points to reduce spatial autocorrelation. Two models, Logistic Regression (LR) and Support Vector Machine (SVM), are constructed, and five-fold cross-validation is used to test model performance through five statistical indicators and AUC values. The results show that, under the specific model configurations and sampling strategy adopted in this study, the LR model achieved higher predictive performance (average test AUC = 0.995) than the SVM model (average test AUC = 0.752) in the comparative assessment. Statistical analysis of the final susceptibility map derived from the LR model indicates that high and very high susceptibility zones account for 80.94% of the total provincial area and contain 87.45% of all recorded geohazard points, which confirms the consistency and reasonableness of the zoning results. Spatially, high-susceptibility areas are concentrated in the western and northwestern loess tablelands, the Fenhe River fault basin, and fault-developed sections of the Lüliang and Taihang Mountains. Thick loess layers, river undercutting and coal mining activities jointly reduce slope stability in these zones. Compared with conventional susceptibility modeling workflows, this study incorporates the RX1day extreme precipitation index and implements Frequency-Ratio-constrained stratified negative-sample selection to reduce training-sample bias. The produced susceptibility maps can provide technical support for differentiated geological hazard risk management, ecological restoration and territorial spatial planning for loess-mountain transition regions in northern China, thereby directly contributing to regional sustainable development and disaster resilience. Full article

To evaluate freeze–thaw damage and asymmetric deterioration in shallow-buried bias tunnel linings in cold regions, an explainable data-driven predictive framework was developed. Guided by the underlying frost heave mechanisms, an optimal support vector machine (SVM) surrogate model was trained using 200 thermal–mechanical coupled samples generated via Latin hypercube sampling (LHS). The model achieved high predictive accuracy, yielding an R² exceeding 0.98 and a root-mean-square error (RMSE) of only 0.036. Using the Shapley additive explanations (SHAP) algorithm to deconstruct damage evolution mechanisms, results revealed that ambient temperature exerts a stable quasi-linear driving effect on localized tensile stress concentrations in the lining, with an equivalent deterioration gradient of approximately 0.0731 MPa/°C. Meanwhile, cumulative freeze–thaw cycles trigger a steady quantitative accumulation of lining structural deformation. Results indicate that prior to material yield, structural deterioration of tunnel linings under severe cold conditions does not manifest as an abrupt nonlinear transition, but rather as a highly stable quasi-linear accumulation. This finding clarifies the safety evolution trajectory during the early- to mid-term service stages of extreme cold tunnel linings, providing a robust quantitative basis for proactive tunnel operational early warning systems. Full article

Background and Objectives: Post-COVID-19 condition (PCC) is a commonly reported disorder that has gained attention from the World Health Organization (WHO). Several studies have examined factors associated with PCC; however, relatively few have combined statistical and machine-learning approaches. Therefore, this study used statistical analysis to identify factors associated with PCC and machine-learning methods to evaluate the relative importance of these factors and their contributions to model predictions of PCC. Materials and Methods: This study employed a cross-sectional observational design in which 963 eligible individuals who had tested positive for COVID-19 were enrolled. Participants were asked about the presence of persistent symptoms lasting for at least 2 months and occurring 3 months after COVID-19 infection, as well as the specific symptoms experienced. The WHO Global COVID-19 Clinical Platform Case Report Form for PCC was used to classify persistent symptoms. Demographic information and medical factors were examined using Poisson regression and machine-learning techniques. Results: A total of 209 (22%) out of 963 reported having PCC with fatigue (45%), followed by bone/joint/muscle pain (34%), one neurological symptom (24%), one pulmonary/respiratory symptom (23%), and one mental health symptom (16%) were the most common persistent symptoms. Modified Poisson regression showed that having exactly two chronic conditions, and experiencing two or more previous COVID-19 infections were significantly associated with the prevalence of PCC. The SHAP beeswarm plot indicated that sex, age, time since last COVID-19 infection, number of chronic conditions, and BMI had the greatest influence on the support vector machine (SVM) predictions. Within the fitted model, female sex, age, a longer time since last COVID-19 infection, the presence of chronic conditions, and higher BMI generally shifted predictions toward the PCC category. Conclusions: Approximately 22% of participants reported persistent symptoms, with fatigue being the most frequently reported, followed by musculoskeletal pain and symptoms affecting other body systems. In the modified Poisson regression analysis, having exactly two chronic conditions and multiple previous COVID-19 infections were significantly associated with higher prevalence of PCC. However, the machine-learning models demonstrated limited discriminative performance. Full article

Driving style, reflected in recurrent patterns of acceleration, braking, speed selection, following distance, gear use, and lane-changing behavior, plays an important role in road safety and is also associated with fuel consumption, emissions, passenger comfort, and vehicle wear. This paper presents a structured review of recent research on driving style analysis, with particular emphasis on its relationship with road-safety outcomes and risk indicators. Following a PRISMA-oriented methodology, studies published between 2015 and 2025 were identified, screened, and synthesized to examine how driving styles are defined, detected, classified, and evaluated. The review shows a clear shift toward data-driven approaches, including feature-based machine learning and representation-learning methods using support vector machines, ensemble models, convolutional neural networks, recurrent neural networks, and hybrid deep learning architectures. Common data sources include smartphone inertial and GNSS signals, CAN/OBD vehicle data, telematics platforms, naturalistic driving datasets, and camera-based perception systems. Safety impact is most often assessed through crashes, near-miss events, traffic conflicts, time-to-collision measures, harsh maneuvers, and composite risk scores. Across the reviewed literature, aggressive and unstable driving patterns are generally associated with reduced safety margins and increased risk, although comparability remains limited by inconsistent label definitions, heterogeneous datasets, indirect safety proxies, and varied validation protocols. The paper also proposes a reproducible database architecture linking drivers, trips, driving events, and safety events to support transparent analysis, benchmark development, and future implementation in fleet monitoring, driver feedback, and connected vehicle applications. Full article

Background: Psychological flourishing is a key indicator of optimal mental health and professional well-being, particularly among psychologists who are routinely exposed to emotionally demanding clinical environments. Although self-compassion has consistently been associated with positive psychological outcomes, few studies have integrated traditional statistical methods with machine learning approaches to predict psychological flourishing among psychologists. Objective: This study aimed to examine the relationship between self-compassion and psychological flourishing among psychologists in Saudi Arabia, identify the unique contribution of self-compassion dimensions, evaluate the predictive performance of supervised machine learning models, and compare their performance with traditional multiple linear regression. Methods: A cross-sectional correlational design was employed, involving 224 psychologists practicing in Saudi Arabia. Participants completed the Self-Compassion Scale and the Flourishing Scale. Descriptive statistics, Pearson’s correlation, and multiple linear regression analyses were performed using IBM SPSS Statistics version 29.0. In addition, Random Forest Regression and Support Vector Regression (SVR) models were implemented in Python using scikit-learn version 1.8.0 to predict psychological flourishing based on self-compassion dimensions together with demographic and professional characteristics. Model performance was evaluated using the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). Results: Overall, self-compassion was positively associated with psychological flourishing (r = 0.627, p < 0.001). Multiple linear regression showed that self-kindness had a significant positive independent association with psychological flourishing (β = 0.292, p = 0.001), whereas over-identification had a significant negative independent association (β = −0.206, p = 0.009). The regression model explained 40.5% of the variance in psychological flourishing (R² = 0.405, p < 0.001). In the held-out test-set comparison using the same predictor set, predictive performance was similar across multiple linear regression (R² = 0.273; RMSE = 3.583; MAE = 2.849), Random Forest Regression (R² = 0.282; RMSE = 3.562; MAE = 2.771), and Support Vector Regression (R² = 0.272; RMSE = 3.587; MAE = 2.768), with no substantial predictive advantage of the machine-learning models over the linear benchmark. Conclusions: Self-compassion, particularly self-kindness and over-identification, was significantly correlated with psychological flourishing among psychologists. Machine-learning models demonstrated predictive performance comparable to traditional regression, with no substantial predictive advantage over the linear benchmark, indicating that increased model complexity did not improve out-of-sample prediction in the present sample. Full article

Classical machine learning (ML) models, including Logistic Regression (LR), Support Vector Machines (SVM), Random Forests (RF), and XGBoost, remain widely used in practical applications because of their efficiency, interpretability, and relatively low computational costs. However, their security properties against different adversarial threats are often evaluated independently rather than within a unified comparative framework. This paper presents a unified empirical evaluation of Logistic Regression, Linear SVM, Random Forest, and XGBoost models across image (MNIST and CIFAR-10), text (AG News), and tabular (Adult, Spambase) classification domains. The models are evaluated under three adversarial scenarios: training-time poisoning, inference-time evasion, and black-box model extraction attacks. The empirical results reveal architecture- and attack-dependent security trade-offs. Random Forest demonstrated relatively greater resilience to random label noise in several configurations but remained vulnerable to targeted poisoning, particularly on AG News. In contrast, Logistic Regression and Linear SVM achieved high black-box extraction fidelity under some active-query configurations, reaching 96.08% and 94.02%, respectively, on MNIST with Q = 10,000. Their evasion results were comparatively interpretable under the evaluated attack procedures, although the observed performance depended on the perturbation budget and optimization configuration. In particular, the higher apparent accuracy of Linear SVM under PGD than under FGSM should be interpreted as a configuration-dependent observation rather than evidence of intrinsic robustness or confirmed gradient masking. Overall, the findings indicate that model robustness depends on the interaction between the attack strategy, data representation, and model architecture. Full article

Sensitive detection of treatment-associated Raman spectral alterations in tumor tissues remains challenging, particularly when such changes are not readily apparent from conventional morphological evaluation. Here, we developed a label-free Raman biosensing strategy combined with explainable machine learning to characterise treatment-associated spectral signatures in melanoma tumours. A B16-F10 melanoma-bearing mouse model was used to compare untreated and PBS-treated controls with cohorts receiving immune checkpoint blockade, anti-angiogenic intervention, or combination therapy. Raman spectra were acquired from multiple spatial regions of melanoma tissues and analyzed using nonlinear dimensionality reduction, supervised classification, and SHAP-based feature interpretation. Although cohort-averaged spectra showed substantial overlap, multivariate analysis revealed treatment-dependent spectral organization, with the combination-treatment cohort showing the most compact and distinguishable spectral profile. Supervised models, including convolutional neural networks, support vector machines, and k-nearest neighbors, further supported the reproducibility of treatment-associated Raman signatures when evaluated using mouse-level validation strategies. SHAP analysis identified discriminative Raman features mainly located within lipid, phospholipid, ester, protein, and collagen-associated vibrational domains, suggesting potential contributions from metabolic- and extracellular-matrix-related biochemical components to treatment-associated spectral discrimination. These findings indicate that Raman spectroscopy integrated with explainable machine learning provides a sensitive, label-free method for distinguishing treatment-associated spectral differences among melanoma tissues. The proposed approach may serve as a complementary spectroscopic tool alongside conventional histological and molecular analyses for investigating treatment-associated tissue-state alterations. Full article

Class imbalance and class cardinality both affect multiclass classification, but their influence on probabilistic estimation has been less explored. This study examines these impacts using an e-commerce dataset. The study follows a four-stage methodology comprising classifier comparative evaluation, controlled class-cardinality analysis, validation using real categorical variables, and class-imbalance evaluation. Multiclass classification tasks are evaluated using Support Vector Machine, Gaussian Naive Bayes, Logistic Regression, Random Forest, and Decision Tree classifiers. Under five-fold cross-validation, performance is assessed using the macro-F1 score, log loss, and accuracy. Results show that macro-F1 score and accuracy decrease as class cardinality increases, causing greater classification difficulty. Tree-based models like Random Forest exhibit more balanced performance across classes. Gaussian Naive Bayes obtains the lowest log loss, indicating more accurate probability estimations. Class cardinality effects are isolated by varying the number of classes while keeping the features and classifier fixed. Increasing class cardinality reduced posterior confidence and increased entropy and log loss. Using real-time categorical variables, these trends are confirmed. Class imbalance primarily affects minority class performance, whereas class cardinality exerts a broader influence on probabilistic confidence and prediction uncertainty. The findings emphasize the necessity to consider class cardinality, class imbalance, and probabilistic metrics when evaluating multiclass classification models. Full article

This study compares a conventional feature-based representation and a physics-guided condition-index (CI) representation for vibration-based fault diagnosis of industrial motors. Both representations were constructed from identical vibration signal segments and evaluated under the same classification conditions using support vector machine (SVM) classifiers. For each representation, a genetic algorithm (GA) was repeatedly applied to the training data to select three representative variables, after which the SVM hyperparameters were optimized using three-fold cross-validation. Permutation Importance and SHapley Additive exPlanations (SHAP) were subsequently used to interpret the contributions of the selected CIs. Independent test motors, whose fault conditions had been established through manufacturer troubleshooting before the present analysis, were excluded from all model-development procedures. For the independent Unbalance and Misalignment test motors, the CI-based model achieved segment-level classification rates of 99.83% and 100%, respectively, whereas the conventional representation showed substantial misclassification. Because these segments originated from a single physical motor for each fault condition, the reported rates represent within-motor segment-level outcomes rather than population-level estimates of diagnostic performance. FFT analysis revealed dominant 1X and 2X components in the corresponding test data, consistent with their established fault conditions. For an additional independent motor identified as Air-gap Unbalance, the CI-based model classified all test segments as Air-gap Unbalance, while the FFT spectrum exhibited fractional-frequency characteristics similar to those observed in the corresponding fault data. Overall, the physics-guided CI representation produced classification outcomes that were more consistent with the established fault conditions and provided a more physically interpretable basis for model decisions under the industrial motor conditions examined in this study. Full article