Abstract. Branched glycerol dialkyl glycerol tetraethers (brGDGTs) are critical molecular biomarkers for the quantitative reconstruction of past environments, ambient temperature, and pH across various archives. However, numerous issues persist that limit their application. The distribution of brGDGTs varies significantly based on provenance, resulting in biases in environmental reconstructions that rely on fractional abundances and derived indices, such as MBT5ME′. This issue is especially significant in shallow lakes, wetlands, and peatlands, where ecosystems are sensitive to diverse environmental and climatic factors. Recent advancements, such as machine learning techniques, have been developed to identify changes in provenance; however, these techniques are insufficient for detecting mixed environments. The probability estimates derived from five machine learning algorithms are employed here to detect provenance changes in brGDGT downcore records and to identify periods of mixed provenance. A new global modern database (n=2031) was compiled to train, validate, test, and apply these algorithms to two sedimentary records. Our findings are corroborated by pollen, non-pollen palynomorphs, and X-ray fluorescence (XRF) obtained from the same sedimentary core sequence. These microfossil and geochemical proxies are utilized to discuss changes in provenance, hydrology, and ecology that influence brGDGT provenance. Probability estimates derived from random forest with a sigmoid calibration are most effective in detecting changes in brGDGT provenance. Minor changes in the relative contributions of brGDGT provenance can significantly influence the distribution of brGDGT, especially regarding the MBT5ME′ index.

ANKRD46 as a shared diagnostic and therapeutic marker in keloid and type 2 diabetes mellitus identified via multi omics and experimental validation.

Understanding the prevalence of genetic dwarfism and developing detection techniques are major difficulties. Genetic dwarfism is defined by below-average stature resulting from genetic alterations. In addition to advances in detection through machine learning algorithms, this abstract investigates the analytical interpretation and comparison of genetic dwarfism statistics. In the first section, we explore the epidemiological context of genetic dwarfism, including prevalence rates, frequencies of genetic mutations, and the range of clinical presentations in various groups. The figures emphasize the intricacy of genetic variants that lead to dwarfism and emphasize the necessity for rigorous analytical methods. Improving detection and diagnostic precision through the use of machine learning algorithms appears to be a potential approach. Machine learning algorithms are trained to identify minor patterns suggestive of genetic dwarfism by utilizing datasets that include genetic profiles, medical histories, and phenotypic features. Effective methods for determining genetic markers and forecasting clinical outcomes related to dwarfism include supervised learning algorithms (e.g., decision trees, support vector machines) and deep learning architectures e.g., Convolutional Neural Networks (CNNs), Generative Adversarial Networks (GANs), Recurrent Neural Networks (RNNs), Autoencoders, Capsule Networks (CapsNets), Graph Convolutional Networks (GCNs), and Long Short-Term Memory (LSTM) networks). A side-by-side comparison highlights the benefits and drawbacks of machine learning techniques over conventional diagnostic techniques. Large-scale genetic data procshines but subtle pattern detection are areas where machine learning shines but deciphering intricate genetic connections and guaranteeing model interpretability in clinical settings continue to be difficult tasks. Moreover, the interdisciplinary aspect of tackling genetic dwarfism with modern computational tools is highlighted by ethical problems pertaining to data privacy, informed consent, and equitable access to genetic testing. Ultimately, this abstract summarizes the state of the art on genetic dwarfism statistics and machine learning applications, promoting ongoing multidisciplinary cooperation to maximize the effectiveness of therapeutic approaches and diagnosis for people with genetic dwarfism.

The growing demand for sustainable and efficient water treatment solutions underscores the importance of high-quality activated carbon (AC) derived from renewable resources. In this study, AC was produced from sugarcane bagasse using sulfuric acid as an activating agent. A hybrid approach combining experimental design and advanced computational modeling was employed to optimize the production process and model the relationship between operational parameters and AC yield. A Box-Behnken design was used to systematically investigate the effects of four key variables: temperature, activation time, raw material-to-activating agent ratio, and acid concentration. The generated experimental data were used to develop and compare predictive models based on response surface methodology (RSM), support vector machine (SVM), and artificial neural networks (ANNs). All models demonstrated strong predictive capabilities, with ANN achieving R=0.989±0.003, outperforming SVM (R=0.950±0.004), while RSM showed a slightly higher overall fit (R=0.996±0.002). This study demonstrates that integrating experimental design with machine learning techniques enhances both the precision and efficiency of process optimization. The proposed approach offers a robust, scalable, and sustainable pathway for producing high-quality AC from agricultural waste, with significant potential for industrial applications in environmental remediation and water purification technologies.

Latent tuberculosis infection (LTBI) is a significant reservoir for active tuberculosis development. Identifying key risk factors is crucial for prevention strategies. Machine learning techniques can uncover complex relationships between risk factors and disease outcomes. Data were collected from China's Tuberculosis Management Information System. LTBI was defined by positive tuberculin skin tests. A case-control design comparing LTBI (n = 669) with active tuberculosis (ATB, n = 669) patients was employed. Propensity score matching (1:1) was performed using age, gender, and education level. Four machine learning models (random forest, XGBoost, support vector machine, and neural network) were developed for feature importance analysis. Least Absolute Shrinkage and Selection Operator (LASSO) regression and logistic regression identified key risk factors. Bootstrap resampling (n = 1,000 iterations) assessed model stability with 95% confidence intervals. Shapley Additive Explanations (SHAP) analysis provided feature importance interpretation. A risk nomogram was constructed and evaluated using receiver operating characteristic curves, calibration plots, and decision curve analysis. Among 1,338 matched participants, XGBoost demonstrated superior performance (AUC = 0.898, accuracy = 85.7%, sensitivity = 84.2%, specificity = 86.9%). SHAP analysis revealed age group (mean |SHAP value|=0.818) as the most influential predictor, followed by medical insurance type (0.599), income group (0.523), and education level (0.439). Logistic regression identified 11 significant risk factors: age (OR = 2.35, 95%CI: 1.86-2.96), BMI (OR = 0.81, 95%CI: 0.71-0.93), smoking status, occupational dust exposure, diabetes, medical insurance type, immunosuppressant use, education level, silicosis, anemia, and TB contact history. The nomogram showed good discrimination (AUC = 0.839) and clinical utility, identifying 64.44% of subjects as high-risk with 53.62% confirmed as true positives at 20% risk threshold. This study successfully identified key LTBI risk factors using machine learning approaches. The developed nomogram provides a practical tool for targeted screening in resource-limited settings. Interventions targeting modifiable factors such as smoking cessation and occupational dust control may reduce LTBI and active TB burden.

According to the data provided by Coface platform, there are almost 3.8 million registered companies in the Visegrad Group (V4), with a significantly increased number of bankruptcies over the last years. Therefore, the main aim of this paper is to identify stable key indicators that determine the financial condition of these companies, which is of crucial importance for stakeholders and investors. To address this topic, we rely on the original dataset consisting of 145,638 company-years from the V4 countries, covering six main sectors during the period of 2018-2021. We calculate 78 financial and non-financial ratios, and we build a robust framework for the identification of the most important ones. Our framework relies on explainable machine learning techniques followed by cross-country and cross-sectional comparisons of the indicators. The results reveal that most of the non-financial indicators included in the analysis are important in assessing the financial condition of companies.

Machine learning is a powerful tool for extracting meaningful patterns from large datasets and performing predictive modeling. In recent years, machine learning methods have been increasingly applied in family sciences, mental health, and educational research. This systematic review aims to evaluate how machine learning methods are used to understand the impact of family dynamics on individuals’ mental health, educational attainment, and behavioral outcomes. A comprehensive literature search was conducted in the Web of Science, PubMed, Scopus, Science Direct, Ulakbim, and TRDizin databases, and 11 studies meeting the PICOS criteria were analyzed. The reviewed studies indicate that machine learning algorithms provide strong predictions in areas such as domestic violence, depression, academic achievement, and children’s psychosocial development. In particular, Random Forest (RF), Support Vector Machines (SVM), deep learning, and natural language processing (NLP) methods have demonstrated high accuracy in predictive tasks. However, challenges related to model transparency, ethical concerns, and applicability within the family context remain among the limitations of machine learning models. Therefore, future research should focus on enhancing the interpretability of machine learning approaches, integrating them with theoretical models, and supporting their application in family sciences with more empirical studies. By doing so, machine learning techniques can be used more effectively to understand family dynamics and support individuals' mental health.

This study evaluates the mechanical performance and predictive modeling of fine-grained soils stabilized with crushed aggregate residue (CAR) or crushed limestone waste (CLW) and Portland cement by integrating the porosity–binder index (η/Civ) and Machine Learning (ML) techniques. Laboratory testing included unconfined compressive strength (qu) and small-strain shear modulus (Go) measurements on mixtures containing 15% and 30% CAR and 3% and 6% cement, compacted at dry unit weights between 1.69 and 1.81 g·cm−3 and cured for 7 and 28 days. Results revealed that strength and stiffness increased significantly with both cement and CAR contents. The mixture with 30% CAR and 6% cement exhibited the highest mechanical performance at 28 days (qu = 1550 kPa and Go = 6790 MPa). When mixtures are compared within the same curing period, the role of CAR and cement becomes evident. At 28 days, increasing CAR from 15% to 30% led to a moderate rise in qu (from 1390 to 1550 kPa) and Go (from 6220 to 6790 MPa). Likewise, at 7 days, increasing cement from 3% to 6% at 15% CAR produced significant gains in qu (207 to 693 kPa) and Go (2090 to 4120 MPa). The porosity–binder index showed strong correlations with qu (R2 = 0.94) and Go (R2 = 0.92). The ML models further improved accuracy, achieving R2 values of 0.99 for qu and 0.97 for Go. Although the index already performed well, the additional gain provided by ML is meaningful because it reduces prediction errors and better captures nonlinear interactions among mixture variables. This results in more reliable estimates for mix design, confirming that the combined use of η/Civ and ML offers a robust framework for predicting the behavior of soil–cement–CAR mixtures.

Alzheimer's disease (AD) remains one of the most challenging neurodegenerative disorders, with limited therapeutic options and high failure rates in clinical trials. This work developed a drug repurposing pipeline powered by a machine learning (ML) model to find possible glycogen synthase kinase-3 beta (GSK-3β) inhibitors, a crucial target in AD pathogenesis. We selected, pre-processed, and optimized a dataset of 4,087 experimentally verified GSK-3β inhibitors using dimensionality reduction and descriptor creation. The most excellent prediction performance was obtained by Random Forest (100 descriptors) out of six supervised ML algorithms that were studied (R2 = 0.8178, RMSE = 0.8118, MAE = 0.6084). Following the virtual screening of 1,616 Food and Drug Administration (FDA)-approved drugs using this refined model, many compounds with projected IC₅₀ < 500 nM were found. Docking experiments showed insightful interactions and high binding affinities with the active-site residues of GSK-3β. With the best docking score (-9.3kcal/mol), stable molecular dynamics (Average RMSD values (1000 ns): protein, 2.23 ± 0.93 Å; protein-ligand complex, 1.40 ± 0.43 Å) and long-lasting contacts with crucial residues, dolutegravir stood out among the top choices. ADMET profiling validated good pharmacokinetics and safety characteristics; however, possible hepatotoxicity needs more research. A HOMO-LUMO gap of 3.07eV was found by density functional theory (DFT) analysis, indicating robust electron transport characteristics and balanced reactivity that are favorable for protein-ligand interaction. Together, these findings show that dolutegravir is a potential repurposable option against AD and how integrative ML, docking, MD, ADMET, and quantum chemistry techniques may speed up the identification of new drugs.

Development of a machine learning-based prediction model for postoperative delirium in frail elderly patients undergoing noncardiac surgery under general anesthesia.

This study employed bioinformatics to identify immune-related diagnostic and therapeutic target genes in thymoma, offering a theoretical foundation for clinical diagnosis and treatment of the disease. In this study, machine learning techniques were employed to identify and validate potential biomarkers, leading to the construction of a diagnostic nomogram. Immune infiltration analysis was subsequently conducted to explore the relationships between these biomarkers and immune cell populations. Finally, the findings were validated through RT-qPCR and Western blotting experiments, ensuring the robustness of the results. Three critical biomarkers were selected for thymoma, with CCR9 and CCL13 confirmed as the final biomarkers through ROC analysis, both demonstrating AUC values exceeding 0.7. Immune infiltration analysis revealed significant differences in 14 types of immune cells between the thymoma and control groups, highlighting a strong association between macrophages and the biomarkers. A nomogram for thymoma was constructed using these 2 biomarkers, exhibiting robust performance. Ultimately, validation experiments using RT-qPCR and Western blotting confirmed the consistency of CCR9 and CCLI3 expression with the bioinformatics results. CCR9 and CCL13 were identified in this study as immune-related biomarkers associated with thymoma, which might provide a theoretical foundation for the development of targeted gene therapies.

Urban waterlogging resilience systems show complex scale effects and cascading dynamics. Accurately identifying their scale structure and operational order strengthens urban waterlogging resilience and upgrades disaster adaptation efficiency. Most existing studies depend on a 'macro-meso-micro' three-tier scale structure. However, their scale divisions are often constrained by data availability or administrative boundaries, neglecting the multiscale, nested characteristics of resilience systems and the alignment between crucial indicators and scale structures. Using Xiamen Island (XMI), China, as a study area, this research creates a comprehensive analytical framework that integrates multi-source data fusion, network percolation models and machine learning techniques to analyse the scale structure of waterlogging resilience systems. The findings reveal that the robustness of the XMI waterlogging resilience system transitions substantially at three key scale nodes, namely 5km², 15km² and 30km², diverging from conventional administrative scales. The impact mechanisms of several resilience indicators on waterlogging adaptation performance depict notable scale dependence and threshold effects, with smaller scales exhibiting more intricate and diverse mechanisms than larger ones. According to XMI's waterlogging resilience scale structure and the non-linear relationships between critical indicators and adaptation results, this study proposes a differentiated indicator management system and dynamic management mechanisms while offering scientific evidence to support the efficient integration of urban spatial management with waterlogging resilience strategies.

This study examines the use of artificial neural networks (ANNs) to classify financial risks in micro-, small-, and medium-sized enterprises (MSMEs) in Montenegro and the wider Western Balkan region. The economies in this region share structural similarities, such as a high concentration of MSMEs, limited access to finance, and vulnerability to macroeconomic volatility, which make financial risk assessment particularly challenging. Traditional statistical and econometric methods often fail to capture the complex, nonlinear interdependencies among financial and operational indicators, resulting in the inaccurate classification of high-risk MSMEs. By applying advanced machine learning (ML) techniques, neural networks (NNs) can identify intricate patterns in multidimensional financial data, significantly improving the accuracy and reliability of risk classification. In this research, a predictive model was developed using key financial and operational variables of MSMEs, enabling the accurate classification of MSMEs in terms of financial instability and insolvency. Empirical validation shows that NNs outperform conventional methods in accuracy, sensitivity, and generalisation. This approach offers tangible benefits for investors, credit institutions, and MSME managers, supporting improvements in early warning systems, optimisation of credit decision-making, and strengthening MSMEs’ financial resilience and sustainability. The methodology also advances risk quantification tools, providing robust indicators for strategic planning and resource management. By focusing the analysis on Montenegro and the Western Balkans, this study demonstrates that regional economic and structural similarities support the adaptation of NN models for precise financial risk classification, offering actionable insights to enhance MSME performance and regional economic stability.

Abstract Star formation (SF) in the interstellar medium (ISM) is fundamental to understanding galaxy evolution and planet formation. However, efforts to develop closed-form analytic expressions that link SF with key influencing physical variables, such as gas density and turbulence, remain challenging. In this work, we leverage recent advancements in machine learning (ML) and use symbolic regression (SR) techniques to produce the first data-driven, ML-discovered analytic expressions for SF using the publicly available FIRE-2 simulation suites. Employing a pipeline based on training the genetic algorithm of SR from an open software package called P y SR, in tandem with a custom loss function and a model selection technique that compares candidate equations to analytic approaches to describing SF, we produce symbolic representations of a predictive model for the star formation rate surface density (Σ SFR ) averaged over both 10 Myr and 100 Myr based on eight extracted variables from FIRE-2 galaxies. The resulting model that PySR finds best describes SF, on both averaging timescales, features equations that incorporate the surface density of gas Σ gas , the velocity dispersion of gas σ gas, z , and the surface density of stars Σ * . Furthermore, we find that the equations found for the longer SFR timescale all converge to a scaling-relation-like equation, all of which also closely capture the intrinsic physical scatter of the data within the Kennicutt–Schmidt plane. This observed convergence to physically interpretable scaling relations at longer SFR timescales demonstrates that our method successfully identifies robust physical relationships rather than fitting to stochastic fluctuations.

The fundamentals of laparoscopic surgery (FLS) program uses box trainers to develop laparoscopic skills. However, these simulators lack personalized training, real-time objective assessment, and primarily represent adult anatomies, neglecting pediatric cases. To address these limitations, advanced objective evaluations like motion analysis and eye-tracking are needed to track trainees' progress and provide real-time formative feedback. However, dynamic training environments challenge eye-tracking data extraction due to shifting areas of interest (AOI). This study aimed to extract AOI-dependent and motion metrics for differentiating and predicting trainees' skill levels across different box trainer anatomies. Medical students and residents performed the peg transfer task on adult and pediatric box trainers. Computer Vision-Deep Learning (CV-DL) algorithms were integrated with eye-tracking data to automatically detect AOIs and extract AOI-dependent (fixation rates on objects and tools) and motion (tool speed) metrics. K-means clustering was used to differentiate trainees' skill levels. To predict trainees' visual behavior, we employed multiple Machine Learning (ML) techniques, including Random Forest, Support Vector Machine, Artificial Neural Networks, and Decision Trees. These methods were used to evaluate which technique could most accurately predict trainees' visual attention patterns. The extracted metrics successfully classified novices into High and Mid-Low skill levels, with significant differences in all extracted metrics between visual behavior levels (p < 0.05). Random Forest achieved the highest accuracy for visual behavior prediction, highlighting the importance of fixation rates on objects and tool speed as key predictors using Gini importance. Results showed consistency in novices' visual attention between pediatric and adult box trainers (p > 0.05). The findings from this work are significant, indicating that novices' skill levels may differ even in their early-stage training, and extracted metrics have the potential to classify and predict novices' skill levels and visual behavior. This is important for customizing and adapting trainees' training programs to enhance their performance.

Indeed, understanding the behavior of headed stud shear connectors in composite steel and concrete construction is essential for ensuring structural integrity and optimal performance. This research focuses on the sensitivity and parametric assessment of the behavior of headed stud shear connectors in composite steel and concrete construction using ensemble machine learning techniques. The study aims to uncover hidden correlations and patterns in the data using a detailed database from 464 push tests, where connectors are welded within the ribs of both trapezoidal and re-entrant steel decks. These patterns provide insights into the performance of shear connectors under various conditions, including different welding methods. The application of ensemble machine learning offers an opportunity to understand complex relationships between variables that may not be immediately evident through conventional analysis. Within the study context, eight types of ensemble machine learning models are implemented and applied to estimate the shear capacity of shear studs and conduct feature importance and partial dependence analysis. The outcomes of this research contribute to a deeper understanding of the factors influencing the performance of shear connectors, providing valuable input for structural design and evaluation in composite construction practices. As a result, this research not only enriches the current academic discourse on shear connectors but also offers pragmatic insights for professionals in the field, thereby bridging the gap between theoretical research and real-world applications in composite construction practices.

The rapid pace of artificial intelligence (AI) and machine learning techniques has necessitated the development of large-scale models that rely on energy-intensive data centers, thereby raising environmental sustainability. Simultaneously, the increasing significance of privacy rights has led to the emergence of Privacy-Preserving Machine Learning (PPML) technologies, which aim to ensure data confidentiality. Although homomorphic encryption (HE) facilitates computations on encrypted data, it entails considerable computational costs and challenges, which impede the effective deployment of privacy-enhancing applications with large models. To create a more sustainable and secure AI world, we propose LIME, a pure HE-based PPML solution, by integrating two techniques: element-wise channel-to-slot packing (ECSP) and power-of-two channel pruning (PCP). ECSP leverages abundant slots to pack multiple samples within ciphertexts, facilitating batch inference. PCP prunes the channels of convolutional layers by powers of two, thereby reducing computational demands and enhancing the packing capabilities of pruned models. Additionally, we implement the ReLU-before-addition block in ResNet to mitigate accuracy degradation caused by approximations with quadratic polynomials. We evaluated LIME using ResNet-20 on CIFAR-10, VGG-11 on CIFAR-100, and ResNet-18 on Tiny-ImageNet. Using the original models, LIME attains up to 2.1% and 8.4% accuracy improvements over the methods of Lee et al. (IEEE ACCESS’21) and AESPA (arXiv:2201.06699), which employ high- and low-degree polynomial ReLU approximations, respectively. Even with 75% parameter pruning, LIME retains higher accuracy than AESPA. Using the state-of-the-art ORION (ASPLOS ’25) as the convolution backend and evaluating on the original models, LIME achieves speedups of 41.5 \(\times\) and 8 \(\times\) over ORION integrated with Lee et al. and AESPA, respectively. For models pruned by 90%, these speedups increase to 202.5 \(\times\) and 35.1 \(\times\) , respectively.

Unveiling the synergistic impact of heavy-metals and alkalinity on microbial diversity using machine learning techniques

In this article, models for prediction of surface hardness for SLM specimens are presented. In experiments, EOS Maraging Steel MS1 was processed using EOS M 290 3D printer via selective laser melting (SLM). To predict hardness of SLM specimens, several machine learning methods were applied, including genetic programming, neural network, multiple regression, k-nearest neighbors, support vector machine, logistic regression, and random forest. In the research, fractal geometry was used to characterize the complexity of SLM-shaped microstructures. It was found that fractal geometry combined with machine learning techniques together greatly improved our comprehension of the intricacies of surface analysis and provided highly efficient predictions. All the applied algorithms exhibited predictability above 90%, with the best average result of 98.7% for genetic programming.

Metabolic dysfunction-associated steatotic liver disease (MASLD) is the most prevalent liver disease globally and a significant public health challenge posing serious threats to human health. Although a liver biopsy is considered the gold standard for diagnosing MASLD, its invasiveness, high cost, and associated risks limit its clinical applicability. With the rapid advancements in artificial intelligence and imaging technologies, non-invasive biomarkers and machine learning techniques are increasingly being used in the diagnosis of MASLD. This article systematically reviews the latest developments in non-invasive diagnostic technologies for MASLD, focusing on four major areas: serological, imaging, multi-omics, and AI-driven diagnostic models. This study aims to provide evidence-based insights for precise clinical diagnosis, ultimately facilitating early warning, dynamic monitoring, and individualized management of MASLD.