Diabetes is one of the global health challenges and requires early detection and an accurate diagnosis for the prevention of serious complications. Traditional methods struggle to handle the complexities of modern data sets. Advanced deep learning techniques can yield better solutions. This paper proposes a novel deep-learning framework optimized for diabetes prediction using the Pima Indian Diabetes Dataset. This is suggested to introduce the CatBoost algorithm and a deep learning architecture involving Convolutional Neural Networks (CNNs) and Bidirectional Long Short-Term Memory (Bi-LSTM) networks. Hyperparameter tuning was performed using the Mountain Gazelle optimizer (MGO) to balance exploration and exploitation in the search space effectively. It achieved the best performance, with an accuracy of 0.955, a precision of 0.96, a recall of 0.95, and an F1-score of 0.95, outperforming traditional algorithms such as Logistic Regression and Naive Bayes, which recorded accuracies of 0.775 and 0.78, respectively. Conversely, this proposed approach outperforms other deep learning methods, including CNNs and Bi-LSTMs, across multiple evaluation metrics, demonstrating strength and potential in clinical diagnostics. This enhances method interpretability; therefore, Recursive Feature Elimination (RFE) is an ideal candidate for medical applications in which clarity in decision-making is crucial.

Scalable data centers – Power generation and delivery challenges and solutions

Ecological niche models (ENMs) are used to assess the abiotic preferences of species by linking their occurrences to the environmental conditions in which they live. We developed a fossil-informed ENM framework that integrates mid-Holocene and modern occurrences to test niche stability and reconstruct abiotic niche characteristics for four critical reef-building Caribbean coral species (elkhorn coral [Acropora palmata], staghorn coral [Acropora cervicornis], boulder brain coral [Colpophyllia natans], and mustard hill coral [Porites astreoides]). Given evidence of niche stability, we used fossil-improved niche estimates to predict area and location of habitat for future climate scenarios in 2050 and 2100. We built species distribution models with environmental predictors and compared models trained with modern-only versus combined fossil and modern occurrences to evaluate differences in niche breadth, model performance, and projected habitat distributions under future climate scenarios. Including mid-Holocene fossil data in ENMs broadened niche estimates, resulting in a larger area of predicted habitat than models based solely on modern data (up to 114,559km2 more in 2100). Although our models showed that suitable habitats existed for most corals in 2100, the amount declined dramatically (45-100% decrease in area from the present day), there was a significant restriction of lower latitude habitat suitability, and marine protected areas did not overlap the majority of predicted future suitable habitat (8-20% overlap by 2100). Fossil-informed models expanded niche estimates in environmental space and incorporated environmental conditions not represented in modern data, resulting in broader projections of future habitat. Our results suggest that actions to reduce emissions and expand protected areas in the northern Caribbean are imperative to prevent significant degradation and that using fossil occurrences in niche estimation can improve the reliability of conservation forecasting, an approach that is transferable across taxa and regions.

Identifying influential variables in high-dimensional matrix-valued data while controlling the false discovery rate (FDR) is a critical challenge in modern data science. We propose a novel, model-free procedure specifically designed for simultaneous row and column selection in matrix predictor regression. Our approach utilizes folding selection subspaces (FSS) to formulate structured hypotheses and employs data splitting to construct mirror statistics from FSS estimators. This design bypasses restrictive model specification and the need for p-value computation. Key theoretical contributions include establishing the asymptotic distribution of FSS estimators and proving the mirror statistic is asymptotically symmetric with respect to zero under the null hypothesis. Using the symmetry, we develop a multiple hypothesis testing procedure with data-driven thresholds that provably controls the FDR for row and column at the desired level asymptotically. The framework is further extended to control element-wise FDR under specific structural assumptions. Extensive simulations and a real data analysis demonstrate the superior performance of the proposed method over existing approaches across various settings.

BackgroundLarge cell lung cancer (LCLC) is an aggressive, undifferentiated subtype of non-small cell lung cancer (N-SCLC) and is now a rare subtype in clinical practice.MethodsData were retrieved from the SEER database, with two analytical cohorts established. Joinpoint regression quantified LCLC incidence trends. Propensity score matching (PSM) balanced baseline characteristics of the survival cohort. Cox regression determined independent overall survival (OS) predictors, restricted cubic spline (RCS) explored non-linear associations between continuous factors and outcomes, and Kaplan-Meier curves with Log-rank tests compared survival differences.ResultsFrom 1992 to 2022, the incidence of LCLC exhibited a significant downward trend (annual percent change [APC] = -12.690%, 95% CI: -13.788 to -11.577, P < 0.001). The most rapid decline was observed during 2005-2015, with an APC of -19.624% (95% CI: -21.955 to -17.224, P < 0.001). Finally, a significant decreasing trend persisted from 2015 to 2022, albeit with a slightly slowed rate (APC = -13.995%, 95% CI: -21.303 to -6.008, P = 0.002). Multivariate analysis identified advanced age, male sex and advanced AJCC stage as independent predictors. Lobectomy and extended lobectomy were associated with improved OS, while no chemotherapy was a risk factor.Conclusions1992-2022 US LCLC incidence decline is attributable to diagnostic drift rather than reduced actual disease burden; our study identified sex, age, AJCC stage, surgical resection extent and chemotherapy as OS predictors for LCLC patients. Notably, SEER lacks modern systemic therapy data, precluding unrigorous extrapolation of its chemoradiotherapy findings to current regimens.

Semantic indexing ensures better organization and optimized searching of heterogeneous, autonomous, and distributed data sources. This approach leverages meaning and context rather than just keywords to better manage the increasing volume, complexity, and heterogeneity of modern data, enabling precise searching, optimized integration, and improved interoperability between domains. Several approaches to semantic indexing are available: ontology-based indexing, machine learning and automated semantic annotation of data sources. However, the main challenge remains scaling up. This article focuses on a conceptual framework designed for scalable semantic indexing of data sources based on a structured peer-to-peer architecture adapted for managing a very large number of nodes, Hilbert curve renowned for its preservation of semantic affinity while scaling, hypercube structure with its efficient diffusion algorithm, semantic annotation of data sources based on keywords, as well as machine learning techniques, in particular, multidimensional data analysis. An illustrative exploratory example of the Meta Skills semantic class is presented to outline the proposed architecture. This study proposes a conceptual and exploratory framework for large-scale semantic indexing of data sources. The proposed approach has not yet been implemented or validated on a large scale; its objective is to provide an initial structured model to serve as a basis for future empirical research.

Abstract The article examines number agreement in the verb olla ‘to be’ in the Soikkola dialect of Ingrian. Although Ingrian distinguishes two grammatical numbers, in the Soikkola dialect there is competition between three verb forms: singular, plural and impersonal. The aim of this study is to identify the factors that determine the choice of a particular verb form in constructions with a plural subject. The analysis is based on recordings made in the 21st century by the authors and their colleagues. The study shows that the most significant factors influencing the choice of the verb form are verb tense and polarity, word order, and individual preferences of a native speaker, which correlate with sub-dialectal zones. A comparison of the results obtained with the data from previously published Soikkola texts suggests that the agreement system observed in the 21st century is not an innovation, because modern data and previous materials exhibit similar patterns.

The Zhuang, China's largest ethnic minority with over 17 million individuals, represent a critical yet understudied population for understanding East Asian genetic diversity and population history. Here, we present the first high-coverage whole-genome (>30×) and exome (>70×) sequencing study of the Zhuang (ZUN), integrating ancient and modern genomic data to reconstruct their evolutionary trajectory. We show that ZUN derive ∼68% of their ancestry from the Tai-Kadai-speaking people, diverging ∼3,000 to 5,000 years ago (ya), with the Maonan as their closest genetic relatives. Our analyses support a shared origin of Tai-Kadai and Austronesian populations ∼7,000 ya, predating their divergence from Sino-Tibetan groups ∼16,000 ya. Substantial gene flow from Han Chinese since ∼4,000 ya reduced genetic divergence between ZUN and northern East Asians to ∼12,000 years. The ZUN ancestral gene pool formed 5,000 to 3,000 ya through multiple admixture waves, with 87% contribution from southern populations and 12% from northern groups. Demographic modeling indicates continuous population expansion until ∼10,000 ya, followed by a pronounced growth surge over the past 400 years. Adaptive selection signatures highlight genes linked to immune response (IGH cluster), lipid metabolism (FADS1/2), wound healing (TMEM121), and environmental adaptation (ABCC11), suggesting dietary shifts and tropical pathogens as key evolutionary drivers. Furthermore, the ZUN genetic profile reflects their role as a regional hub for gene flow into neighboring populations, coinciding with Han migrations during the Qin dynasty. Together, these results identify the Zhuang as descendants of Baiyue populations with a distinctive dual ancestry shaped by Neolithic southern and northern East Asians.

Managing a vast number of virtual machines (VMs) efficiently is a critical challenge in modern large-scale data centers. The continuous creation and termination of VMs lead to resource fragmentation across physical machines (PMs), necessitating periodic VM rescheduling to optimize resource utilization. Despite its significance, VM rescheduling has received limited attention in the literature. A key challenge is that, unlike conventional combinatorial optimization problems, the efficiency of rescheduling algorithms is heavily impacted by inference time, as VM states evolve dynamically during execution. This scalability bottleneck hampers existing methods. To address this, we propose VMR <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> L, a reinforcement learning framework tailored for VM rescheduling. VMR <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> L integrates a two-stage decision-making process to accommodate complex operational constraints, a feature extraction mechanism that captures critical relational information for rescheduling, and a risk-aware evaluation strategy that enables users to balance execution speed and rescheduling accuracy. Extensive experiments using real-world data from a production-scale data center demonstrate that VMR <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> L achieves near-optimal performance while reducing inference time to a matter of seconds. To facilitate reproducibility, we provide access to our implementation and datasets.

Over the past several centuries, the Baltic Proper harbour porpoise (Phocoena phocoena) has undergone significant population declines, resulting in its current IUCN classification as critically endangered. While conservation efforts are extensive and multinational, the population's abundance has only been estimated once (SAMBAH Project, 2011-2013). Ad hoc historical sources and the sub-fossil archaeological record, dating to 7000 years BP, suggest that the population once had a wider distribution in the Baltic Sea. However, the historical abundance and distribution of the Baltic Proper harbour porpoise population remain largely unknown, especially before the mid-20th century. This study examines archival Swedish newspaper records from the late 1700s to the early 1900s to assess the presence and distribution of porpoises. Digitized articles were searched for by keyword in the National Library of Sweden Newspaper Archive. This dataset was combined with HELCOM/ASCOBANS historic data. The records show that harbour porpoises historically occurred regularly along the entire Swedish coast, including the northernmost parts of the Gulf of Bothnia, where the species is virtually absent today, suggesting a notable range retraction. While harbour porpoises appeared less frequently in the Baltic Sea than along the west coast of Sweden, the number of Baltic records indicates that porpoises occurred more frequently than today. The peak occurrence of porpoises during spring and summer suggests that the Gulf of Bothnia historically constituted an important foraging habitat for animals migrating from the southern Baltic. By integrating historical records with modern conservation data, this study provides critical insights into the long-term effects of human activity on the Baltic Proper harbour porpoise. Understanding past ranges and approximate population size is vital for guiding more effective conservation strategies.

This article analyzes the main approaches to measuring inflation in developed countries and develops methodological implications for Uzbekistan. Based on the experience of the United States, the European Union, the United Kingdom, Japan, and Australia, the study examines systems of consumer price indices, core inflation indicators, weighting and aggregation methods, as well as emerging trends in the use of digital data sources. The results show that in developed economies, inflation measurement is not limited to a single aggregate index; rather, it is based on a comprehensive system of indicators designed for different purposes, supported by modern data sources and continuous methodological updates.

The rapid growth of online transactions and digital banking has significantly increased the use of credit cards for financial activities. However, this convenience has also led to a rise in credit card fraud, causing major financial losses for banks and customers. Traditional fraud detection systems mainly rely on rule-based methods, which are often inefficient in identifying new and complex fraud patterns. To address this challenge, this project proposes a Machine Learning Based Credit Card Fraud Detection System that automatically analyzes transaction data and identifies fraudulent activities with high accuracy.The system utilizes modern technologies such as Python for backend development and machine learning algorithms to analyze transaction patterns and detect anomalies. Transaction data is first preprocessed to remove noise and extract important features such as transaction amount, time, and location. Machine learning algorithms like Logistic Regression, Random Forest, and Isolation Forest are then applied to classify transactions as legitimate or fraudulent. The proposed system aims to improve fraud detection accuracy, reduce financial losses, and enhance the security of online transactions. By leveraging data-driven techniques and intelligent algorithms, the system can identify suspicious activities in real time and assist financial institutions in preventing fraudulent transactions. Experimental results show that the machine learning model can effectively detect fraudulent behavior and improve the efficiency of fraud detection systems.Furthermore, the system provides a scalable architecture for handling large volumes of transaction data and supports faster decision-making in financial security systems. By integrating machine learning with modern data processing techniques, the proposed system offers a reliable and efficient solution for credit card fraud detection in digital payment environments.

Cloud storage systems play a vital role in modern data management; however, they face significant security challenges such as unauthorized access, insecure key exchange, and lack of data integrity verification. This paper proposes a secure cloud data storage system using hybrid encryption. The system combines AES-256 GCM for efficient file encryption and RSA for secure key exchange. SHA-256 hashing is used to ensure data integrity. Additionally, role-based access control (RBAC) and activity logging mechanisms are implemented to enhance system security. Experimental results demonstrate improved confidentiality, key protection, and controlled access compared to traditional cloud storage systems.

Predictive modeling is an essential component of modern data science, driving decision-making across domains such as healthcare, business, and engineering. One of the primary challenges in building reliable models is identifying correlations among features and selecting the most influential variables. This paper presents an interactive web application that automates and evaluates feature importance using machine learning techniques. The platform integrates a Flask backend with libraries such as Pandas, Scikit-learn, Matplotlib, Seaborn, Plotly, XGBoost, and SHAP to deliver real-time data analysis, interactive visualizations, and model interpretability. The system supports Pearson and Spearman correlation, Random Forest and permutation-based feature importance, and optional model training using decision trees, regression models, and ensemble methods. Experiments conducted on datasets from Kaggle and UCI repositories show that the platform reduces analysis time by 95% and increases model accuracy by up to 14% when compared to traditional manual workflows.

The increasing reliance on data-driven decision- makingdemandsintelligentsystemsthatcanautonomouslyplan, govern, and enforce data processes while ensuring compliance and scalability. This paper surveys advancements in multi-agent platforms and introduces a conceptual framework comprising fourspecializedagents:aTaskPlannerAgent,aDataGovernance Agent,aDataEnforcementAgent,andaSyntheticDataCreation Agent. Leveraging Large Language Models (LLMs) where appropriate, the Task Planner Agent interprets natural language input to generate actionable roadmaps, enabling automated orchestrationofcomplexworkflows.TheDataGovernanceAgent oversees adherence to policies and regulatory standards, while the Data Enforcement Agent ensures their execution in realtime, safeguarding integrity and compliance. The Synthetic Data Creation Agent generates privacy-preserving data to support ex- perimentation and model development. Collectively, these agents establish an LLM-enhanced platform designed to address plan- ning,governance,enforcement, anddataaugmentation challenges in modern data ecosystems

Background. Many pharmaceuticals, including antibiotics, diuretics, some antitumor agents, hormones, etc., can promote the depletion of magnesium (Mg), pyridoxine (vitamin B6, VB6), and other micronutrients (MNs) in the body. This process may lead to the development of hypomagnesemia and concomitant MN deficiencies, which are associated with a range of adverse effects, including neurotoxicity, cardiotoxicity, hepatotoxicity, etc. Moreover, the resulting micronutrient deficiency (MND) may paradoxically aggravate the underlying pathophysiological mechanisms of the diseases for which these drugs are prescribed, thereby potentially diminishing therapeutic efficacy and contributing to treatment-related complication. Objective: Chemoreactomic assessment of anti-micronutrient (anti-MN) effects of all drugs included in the Anatomical Therapeutic Chemical (ATC) classification system. Material and methods. Using modern data mining techniques, including mathematical approaches from topological data analysis, labeled graph theory (chemographs), and related method, this study performed a systematic computer-based analysis of databases describing the Mg-depleting effects of drugs; original algorithms for numerically predicting the Mg- and VB6-removing effects of drugs. Original algorithms were developed for the numerical prediction of Mg- and VB6-depleting properties of drugs, as well as for the assessment of other anti-MN effects. These algorithms were subsequently applied in a chemoreactomic screening of 2,527 drugs classified within the ATC system. Results. A database describing anti-MN properties of drugs was created for 24 MN balance indicators for 18 MNs. Algorithms for predicting the anti-MN properties of drugs were developed with a classification accuracy of 92±10% in cross-validation (the accuracy of predicting VB6 MND – 88%, Mg MND – 94-98%). On average, each drug from the ATC group accounts for 8.5±6.5 anti-MN effects. Only 100 out of 2527 (4%) drugs did not exhibit a negative impact on MN, primarily amino acids, MNs themselves, and choline drugs. The most pronounced negative impact of the drugs under study was related to the metabolism of vitamin D3 (505 ATC categories), VB6 (475 ATC categories), iron (419 ATC categories), vitamin B1 (386 ATC categories), and Mg (375 ATC categories). VB6 MND was caused by 1701 drugs, Mg MND – by 1064 drugs. Antibiotics for systemic use (ATC code J01), psycholeptics (N05) and psychoanaleptics (N06), antineoplastic agents (L01), sex hormones and modulators of the reproductive system (G03), analgesics (N02), antidepressants (N06A), diuretics (C03), antihistamines for systemic use (R06A), anti-inflammatory and antirheumatic agents (M01), direct-acting antivirals (J05A), and antiepileptic agents (N03A) were found to affect adversely the homeostasis of both Mg and VB6. A detailed description of the anti-Mg and anti-VB6 properties of these drug classes was provided. The data obtained via chemoreactomic analysis were compared with that obtained by experimental and clinical studies of Mg and VB6 preparations. Conclusion. The conducted chemoreactomic analysis provides a substantiated basis for supporting pharmacotherapy with selected medicinal preparations based on organic salts of Mg and VB6.

Advances in organoid and other three-dimensional culture systems, single-cell and spatial transcriptomics, multi-omics, and high-resolution imaging are reshaping our understanding of the cellular origins and evolutionary trajectories of glioblastoma. When integrated with modern data science approaches, these technologies enable the construction of increasingly detailed molecular biographies of normal neural stem and progenitor cells as well as malignant stem-like cellular states. Such molecular biographies illuminate how developmental programs, cellular plasticity, and microenvironmental cues are co-opted during gliomagenesis. At the same time, progress in machine learning, immunotherapy, and precision molecular targeting is beginning to translate these biological insights into therapeutic strategies that specifically disrupt glioblastoma stem-like states. Together, these converging approaches provide a conceptual and technological framework for improved tumor modeling, earlier detection, and increasingly personalized therapies for malignant gliomas.

The rapid expansion of digital platforms has led to an unprecedented increase in user-generated textual content, particularly in the form of customer reviews, which serve as a primary medium for expressing consumer opinions. A notable characteristic of this modern data is its inherent multilingualism and the phenomenon of code-switching, which creates critical bottlenecks for traditional monolingual sentiment analysis tools. This research proposes an automated, high-precision framework for Multilingual Sentiment Analysis utilising a Hybrid Bidirectional Long Short-Term Memory (BILSTM) architecture. The study employs a large-scale dataset consisting of 50,000 multilingual reviews from e-commerce platforms. To ensure data integrity, a rigorous 12-step computational pipeline was established, involving the removal of 418 duplicate entries and the application of a custom text transformation function. Advanced preprocessing techniques, including Regex-based tokenisation, Porter Stemming, and script normalisation, were implemented to reduce linguistic noise and consolidate the vocabulary into a 10,000-word index. The core methodology involves mapping these cleaned tokens into a 128-dimensional dense vector space to achieve language-agnostic semantic alignment. The architectural framework utilises stacked BILSTM layers with 128 and 64 units, respectively, optimised via the Adam algorithm and protected against overfitting through Spatial Dropout (0.3) and Early Stopping. Experimental results demonstrate that the proposed BILSTM model achieved a superior test accuracy of 87.82%, significantly outperforming the standard unidirectional LSTM baseline of 85.57%. The model correctly identified 4,334 negative and 4,152 positive reviews from a test subset of 9,917 samples, proving its high discriminative power. Furthermore, the bidirectional gates proved highly effective at capturing sentiment intent in extended reviews reaching lengths of up to 2,525 words. This study concludes that the integrated deep learning pipeline effectively transforms raw, chaotic user feedback into structured, actionable insights, providing a scalable solution for real-time global consumer expression analysis in diverse linguistic environments.

Time-series clustering is a prominent research area with extensive practical applications. Given the complexity and diversity of modern time-series data, this study proposes a novel time-series clustering method based on inter-network heterogeneity. First, each time-series is converted into a network by using two types of time-series segmentation techniques. Second, an inter-network clustering approach based on graph spectral theory is introduced: we calculate the total variation (TV) distance between the empirical spectral distributions of each network and identify distinct clusters using a hierarchical clustering algorithm. From the perspective of symmetry, networks constructed from similar time-series tend to exhibit comparable spectral structures, which reflect the underlying structural symmetries of their dynamics. Differences in spectral distributions correspond to symmetry breaking among networks, providing an effective mechanism for distinguishing heterogeneous time-series patterns. Our method effectively preserves more distinctive features inherent in the original time-series. To evaluate the performance of the proposed method, simulation studies are conducted, including the recognition of both stationary and non-stationary sequences. The method also performs well on real-world datasets, such as stock closing prices. These results demonstrate that our approach can handle non-stationary sequences and identify the intrinsic correlations in time-series.

Healthcare institutions worldwide generate growing volumes of heterogeneous clinical data, yet legal, ethical, and infrastructural constraints often prevent these data from being centralized for analysis. Federated learning approaches offer a promising solution by enabling multi-site computation without transferring sensitive patient information, but require well-designed cross-site data harmonization. Modern Data Lakehouse architectures address this requirement by providing a scalable, governed foundation for multimodal clinical dataset integration through unified storage, metadata-rich governance, and FAIR-aligned data access. Despite increasing interest in such technologies across the Middle East and North Africa (MENA) region, operational deployments remain limited due to fragmented infrastructures, insufficient data governance, and gaps in practical expertise. This perspective article reports on a German-Tunisian knowledge and technology transfer initiative conducted within the DAAD Ta'ziz Partnership programme. As mentioned in the, 'the Arabic word 'Ta'ziz' means 'strengthening/consolidation' and has been chosen to clearly express the intended outcome of the programme' [https://www.daad.de/en/information-services-for-higher-education-institutions/further-information-on-daad-programmes/taziz-partnership/ (visited on November 28th, 2025)]. The collaboration between the University Hospital of Cologne and the University of Sfax introduced and implemented federated learning concepts via the Personal Health Train paradigm, and explored the design of a Data Lakehouse tailored to emerging healthcare ecosystems in Tunisia. Through an internship programme, hands-on MLOps training, and a large-scale workshop, the project built technical capacity in containerized analytics workflows, data governance, FAIR data management, and lakehouse engineering. We synthesize lessons learned regarding infrastructural limitations, data governance maturity, interoperability challenges, and institutional readiness, and outline considerations for sustainable adoption of distributed analytics in the MENA region. The findings highlight the critical importance of capacity building, bidirectional knowledge exchange, proof-of-concept validation, and administrative engagement for deploying trustworthy AI and modern data infrastructures in sensitive healthcare environments. We by emphasizing the need for further developments regarding federated learning and Data Lakehouse adoption in Tunisia, and how cross-regional partnerships can accelerate responsible, privacy-preserving digital health innovation.