Abstract Athlete fatigue and overtraining are critical factors affecting performance and health, yet traditional evaluation methods relying on subjective judgment or single-indicator monitoring lack systematic and real-time capability. This study proposes a novel Meta-Learning Ensemble Framework (MLEF) integrating multidimensional physiological monitoring for intelligent fatigue risk prediction. The MLEF architecture consists of three progressive layers: a Feature Selection Layer using ANOVA F-statistic based univariate selection to identify the top 12 features from 15 original variables, a Base Learner Layer training four heterogeneous logistic regression classifiers with different regularization configurations, and a Meta-Learning Layer integrating predictions through weighted voting and stacking ensemble strategies. We constructed experiments on the AFR-1000 dataset containing 1000 athletes with balanced class distribution (51:49 normal/fatigue), split 8:2 into training and testing sets with stratified sampling. On the independent test set, MLEF achieved 99.00% accuracy, 98.98% F1-score, and 99.89% ROC-AUC, significantly outperforming traditional machine learning methods (Logistic Regression 98.50%, SVM 92.50%, XGBoost 88.00%) and deep learning models (Attention Network 97.50%, DNN 97.50%). Ablation experiments demonstrated that ANOVA F-statistic based feature selection maintained baseline performance while reducing dimensionality, and progressive ensemble integration raised F1-score from 98.46% to 98.98%. SHAP interpretability analysis identified HRV (mean |SHAP $$|=3.95$$ | = 3.95 ), HeartRate_Recovery (2.89), and Cortisol_Level (2.46) as top predictors, with HRV-Lactate interaction revealing synergistic amplification of fatigue risk. The MLEF model provides a practical AI tool for training monitoring with high accuracy and interpretability, offering scientific guidance for personalized training and recovery planning.

Rapid discrimination of Artemisia Argyi Folium origin using three-dimensional fluorescence combined with Chemometrics.

Limited experimental data remains a key challenge in applying machine learning to drug discovery, particularly for cancer-related targets. In this study, we present a data-efficient active meta-deep learning framework to predict mitogen-activated protein kinase 1 (MAPK1) inhibitors, which are promising candidates for cancer-related therapies. Our approach integrates active learning (AL) with a meta-model that combines four deep architectures: a convolutional neural network, an attention, a graph convolutional network, and a graph neural network-attention, trained on molecular descriptors and graph-based representations. These models generate four probability-based features that feed into an attention-based meta-learner, improving predictive performance by 5.12% in the area under the precision-recall curve (AUPRC) and 5.48% in the Matthews correlation coefficient (MCC) using only 10% of the training data. Among the AL sampling strategies evaluated, entropy sampling showed competitive performance in selecting informative molecules for model improvement. Overall, our framework achieves an AUPRC of 0.835 ± 0.017 and MCC of 0.817 ± 0.017, on par with a traditional training method despite using only 26.7% of the training data. Compared to a conventional random forest model trained on brute-force, a 100% full training set, our approach shows a 10.6% improvement in AUPRC and modest gains in MCC, confirming the effectiveness of the proposed framework. Under severe class imbalance, balanced accuracy steadily increased across AL iterations, reaching values greater than 0.85 at the final iteration for all uncertainty-driven strategies. Molecular docking confirmed successful prioritization of the top four predicted compounds. Evaluation on an external MAPK1 data set demonstrated generalizability, with our approach achieving an AUPRC of 0.818 and an MCC of 0.403, comparable to the independent test set. These results highlight the potential of combining intelligent data selection with deep learning architectures through the meta-model to accelerate predictive performance in data-scarce drug discovery. Scientific contribution: This study contributes a novel, data-efficient active meta-deep learning framework for predicting MAPK1 inhibitors, addressing the challenge of limited experimental data in a cancer-specific target. By integrating AL with a meta-model composed of four deep architectures, the approach significantly enhances the predictive performance using only a fraction of the training data. The framework achieves superior metrics compared to traditional training methods, highlighting its potential to accelerate drug discovery in data-scarce settings.

Do individuals possess a "gaze fingerprint" that reveals how they uniquely look at the world? We tested this question by examining intra- and intersubject gaze similarity across 700 static pictures of complex natural scenes. Independent discovery (n = 105) and replication data sets (n = 46) of adults aged 18 to 50 years (sampled from Italy and Germany) revealed that gaze fingerprinting is possible at relatively high rates (e.g., 52%-63%) compared with chance (e.g., 1%-2%). We also identify gaze-fingerprint barcodes, which reveal a unique individualized code describing which stimuli an individual can be gaze-fingerprinted on. Preregistered longitudinal follow-up experiments have shown that gaze-fingerprint barcodes are nonrandom within an individual over short and long time fraframmes. Finally, we find that increased gaze fingerprintability for social stimuli is associated with decreased levels of autistic traits. To summarize, this work showcases the potential of gaze fingerprinting for isolating traitlike factors that may be of high neurodevelopmental and biological significance.

By influencing gene expression and contributing to epigenetic modifications, Ribonucleic Acid (RNA) 5-Hydroxymethylcytosine (5hmC) modification significantly affects cellular pathways. It plays an important role in complex regulatory networks and gene expression. Moreover, 5hmC modifications are linked to a variety of human diseases, including diabetes, cancer, and cardiovascular conditions. However, experimental methods to identify RNA 5hmC modifications, such as chromatography and Polymerase Chain Reaction (PCR) amplification, are costly and time-consuming. So, computational methods are necessary to predict these modifications. In this study, several feature descriptors were analyzed and compared to finalize the best ones. Different deep-learning models were explored to design the proposed model architecture. Neighbourhood analysis was conducted on the dataset to provide insights into a deeper understanding of RNA 5hmC modifications. The proposed model, InTrans-RNA5hmC, is a dual-branch deep learning model that has two branches: the Inception branch and the Transformer branch. Word embeddings having the contextual information and language model embeddings from the RiboNucleic Acid Language Model (RiNALMo) were used as the finalized feature descriptors. InTrans-RNA5hmC outperformed existing SOTA methods, achieving 0.97 sensitivity, 0.985 balanced accuracy, and 0.985 F1 score on the Independent test set.

Quantitative Evaluation of Artificial Intelligence-Based Organ Segmentation Across Multiple Anatomic Sites Using 8 Commercial Software Platforms.

FEAOF: A transferable framework applied to prediction of hERG-related cardiotoxicity.

circIRES-DAF: A dual-attenuation fusion framework for identification of internal ribosome entry sites in circular RNAs.

Transformer-based feature extraction approach for hematopoietic cancer subtype classification.

An interpretable model based on weakly supervised learning for uterine smooth muscle tumor diagnosis: A multi-center study.

Early identification of pediatric depression in western China: A multicenter, citywide evaluation of nine machine learning models.

Multiperspective tumor heterogeneity metrics for preoperative prediction of IASLC grading in clinical stage IA lung adenocarcinomas: A multicenter study.

An integrated machine learning framework for predicting anthropogenic and natural iodine isotopes in the South China Sea with uncertainty quantification.

Independent set reconfiguration under bounded-hop token jumping

On the number of independent sets in Halin graphs

Magnetic resonance imaging (MRI) radiomics has shown promise in glioma grading and isocitrate dehydrogenase (IDH) mutation prediction, but traditional whole-tumor approaches overlook intratumoral heterogeneity, limiting diagnostic accuracy and interpretability. This study aims to explore cellularity habitat-based MRI radiomics for precise grading and IDH mutation status prediction in adult-type diffuse glioma (ADG). A total of 625 ADG patients were retrospectively collected. Whole-tumor volumes of interest (VOIs) were delineated on four conventional MRI sequences (T1WI, T2WI, T2-FLAIR, and CE-T1WI) and segmented into three cellularity habitats using apparent diffusion coefficient (ADC)-based K-means clustering: H1 (low ADC), H2 (medium ADC), and H3 (high ADC). Radiomic features were extracted from individual and combined habitats, and predictive models were developed using a disentangled-learning-based multi-sequence fusion network (DMSFN). Performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy (ACC), sensitivity (SEN), and specificity (SPE). The optimal habitats for ADG grading (Grade 2 vs. Grade 3 + 4, Grade 2 + 3 vs. Grade 4) and IDH prediction were H1 + 2, H1 + 2, and H2 + 3, respectively. Combining T1WI, CE-T1WI, and T2-FLAIR sequences yielded the highest AUCs of 0.9360, 0.9605, and 0.8721 in the training set, and 0.8070, 0.8236, and 0.8180 in the independent test set. Shapley Additive exPlanation (SHAP) analysis identified key radiomic features contributing to model predictions, with CE-T1WI features consistently demonstrating high discriminative power. Integrating ADC-derived cellularity habitats with MRI radiomics significantly improves the accuracy and biological interpretability of ADG grading and IDH mutation status prediction, offering a robust, non-invasive approach for glioma characterization. Retrospectively registered.

The optimal target of mean arterial pressure (MAP) remains controversial in sepsis management. Critical closing pressure (Pcc), the arterial pressure at which blood flow ceases, is the key determinant of vascular waterfall phenomenon. Tissue perfusion pressure (TPP), the difference between MAP and Pcc, represents the driving pressure for arterial blood flow. This study evaluated the prognostic value of Pcc and TPP for improving risk stratification in sepsis. This retrospective cohort study included adult patients with sepsis in 18 hospitals between August 2013 to October 2022 from two independent data sets (the Study on the Epidemiology, Diagnosis and Treatment of Sepsis [SEPSIS-EDT] registry and the critical care database of Peking Union Medical College Hospital, Beijing, China). Pcc was estimated via linear regression of hourly MAP against product of heart rate and pulse pressure, while TPP was calculated as MAP minus Pcc. Patients were categorized into four groups based on the optimal thresholds for mean Pcc and TPP within 24 h of sepsis diagnosis: low TPP-low Pcc, low TPP-high Pcc, high TPP-low Pcc, and high TPP-high Pcc. Clinical outcomes included mortality rates and development of acute kidney injury within 2 and 7 days of sepsis diagnosis. External validation was performed using the Medical Information Mart for Intensive Care IV (MIMIC-IV) cohort. A total of 6,769 patients (mean age, 61 yr; 61.0% men) were included. Intensive care unit mortality was highest in the low TPP-low Pcc group and lowest in the high TPP-high Pcc group (35.1% vs . 20.1%; risk difference, 15.0%; 95% CI, 10.2 to 19.8%). Similar patterns were observed for other outcomes. After adjustment for MAP, increased Pcc with concomitant reduced TPP showed a significant U-shaped association with both mortality and acute kidney injury development ( P < 0.001). The findings were consistent in the MIMIC-IV cohort. While MAP remains central to sepsis management, Pcc and TPP provide complementary prognostic information. Incorporating these parameters into clinical assessment may improve risk stratification and optimize blood pressure management.

Objective This study aimed to develop an explainable fusion model that integrates intratumoral, peritumoral, and habitat features derived from MRI to evaluate its feasibility for predicting the WHO/ISUP nuclear grade of clear cell renal cell carcinoma (ccRCC). Methods We retrospectively enrolled 154 patients with pathologically confirmed ccRCC and partitioned them into a training set (n = 108) and an independent test set (n = 46). On contrast-enhanced T1-weighted images, regions of interest were manually delineated layer-by-layer along the tumor margin and expanded outward by 1 mm, 2 mm, 3 mm, 4 mm and 5 mm to derive peritumoral regions. Tumor habitat regions were identified using the K-means clustering algorithm. After extraction and selection of radiomic features, radiomics and habitat models were constructed using five machine learning algorithms. These effective features were then integrated into a nomogram. Model performance was assessed by plotting receiver operating characteristic (ROC) curves and calculating the area under the curve (AUC). Model calibration and clinical utility were evaluated using calibration curves and decision curve analysis (DCA). Model interpretability was enhanced by employing Shapley Additive exPlanations (SHAP). Results Three habitat subregions were identified within tumors. The integrated habitat region(Habitat) model demonstrated the highest performance among the evaluated habitat models, with AUCs of 0.894 and 0.877 in the training and test sets, respectively. The Peri2mm model achieved AUCs of 0.884 and 0.839, outperforming other peritumoral ranges. Therefore, the 2-mm peritumoral margin was considered a potentially optimal analysis range in this cohort.When the integrated habitat region signature was combined with intratumoral features, 2-mm peritumoral features and the independent clinical predictor (corticomedullary enhancement level) in a nomogram, predictive performance was further improved, achieving AUCs of 0.934 and 0.912. SHAP bee swarm and force plots provided intuitive visualization of the habitat model’s decision-making process. Conclusion The nomogram, which integrates intratumoral, peritumoral and habitat radiomic features derived from MRI, demonstrated excellent performance for noninvasive preoperative prediction of WHO/ISUP nuclear grade in ccRCC and holds promise as an adjunctive tool for individualized therapy planning and prognostic assessment. However, its clinical application requires further external validation.

Primary immunodeficiencies (PIDs) are rare disorders caused by immune system defects that are commonly screened using multi-parameter flow cytometry (FCM). To counter the subjective and time-consuming manual data analysis of FCM data, we present PIDgeon, a fully automated computational pipeline based on artificial intelligence (AI) techniques. PIDgeon is designed to characterize PID immune profiles, suggest PID subtypes based on altered immune profiles, age, and immunoglobulin levels, and generate interpretable reports. The PIDgeon pipeline, including FlowSOM and extreme gradient boosting models, was trained and tested on standardized FCM data generated according to EuroFlow procedures on 74 healthy controls and 399 patients (281 lymphoid-PID patients and 118 non-PID diseased controls) collected by the Ghent University Hospital. Subsequently, multi-centric validation was performed on internal (n = 211) and external (n = 338) independent data sets collected across 4 EuroFlow centers. Validation demonstrated high accuracy in cell count enumeration, achieving correlation scores above 0.90 for the major lymphocyte subsets. Interestingly, PIDgeon showed high sensitivity (93% to 100%) in predicting PID with severe T-cell defects, such as severe combined immunodeficiency and late-onset combined immunodeficiency, and low false-negative rates (1.5% to 5.4%) for distinguishing other lymphoid-PID vs non-PID diseased controls across data sets. Additionally, PIDgeon gives a first hint toward prediction of subtypes of primary antibody deficiencies, such as common variable immunodeficiency. In summary, PIDgeon is an accessible and explainable AI-pipeline aligned with current clinical needs, aiding laboratory immunologists in early PID diagnostics and increasing data analysis efficiency.

It is a challenge to identify food waste sources in all-weather industrial environments, as variable lighting conditions can compromise the effectiveness of visual recognition models. This study proposes and validates a robust, interpretable, and adaptive multimodal logic fusion method in which sensor dominance is dynamically assigned based on real-time illuminance intensity. The method comprises two foundational components: (1) a lightweight MobileNetV3 + EMA model for image recognition; and (2) an audio model employing Fast Fourier Transform (FFT) for feature extraction and Support Vector Machine (SVM) for classification. The key contribution of this system lies in its environment-aware conditional logic. The image model MobileNetV3 + EMA achieves an accuracy of 99.46% within the optimal brightness range (120–240 cd m−2), significantly outperforming the audio model. However, its performance degrades significantly outside the optimal range, while the audio model maintains an illumination-independent accuracy of 0.80, a recall of 0.78, and an F1 score of 0.80. When light intensity falls below the threshold of 84 cd m−2, the audio recognition results take precedence. This strategy ensures robust classification accuracy under variable environmental conditions, preventing model failure. Validated on an independent test set, the fusion method achieves an overall accuracy of 90.25%, providing an interpretable and resilient solution for real-world industrial deployment.