Random Forest Model Research Articles

Integration of Machine Learning and VIRS for Soil Potassium Content Prediction

ABSTRACT At present, the combination of machine learning and visible and infrared spectroscopy (VIRS) stands out as one of the most promising methods for predicting soil potassium content. In this study, a total of 100 soil samples were collected in the vicinity of Xunsi River, Hubei University of Technology. Spectral information and corresponding soil potassium content were obtained for these 100 soil samples. Subsequently, the original spectra of the soil samples were denoised using the locally weighted regression (loess) method. Feature wavelength selection was performed using ∆Gini and Successive Projections Algorithm (SPA) for dimensionality reduction. The model was then optimized by adjusting its parameters. Finally, random forest (RF), back propagation (BP), Genetic algorithm-back propagation (GA-BP) and particle swarm optimization-back propagation (PSO-BP) prediction models were established to predict soil potassium content. The results show that the RF model exhibits superior prediction performance compared to other algorithms, with a coefficient of determination (R2) of 0.851 and a root mean square error (RMSE) of 0.503. The RF model accurately predicts soil potassium content and validates the stability of the VIRS method for rapid estimation. Among the BP model and its optimization variants, the PSO-BP-Levenberg-Marquardt model demonstrates the best predictive capability, yielding an R2 of 0.791 and an RMSE of 0.621. This combination algorithm holds potential as a valuable reference for the selection of optimization algorithms and training functions for the BP model within the research domain.

Effects of geographical and soil factors on soilś arsenic levels: a case study in typical arsenic-contaminated paddy fields based on machine learning.

Heavy metal pollution in agricultural land has emerged as a contemporary environmental issue of prominent concern. The concentration of heavy metals in soil is influenced not only by inherent soil properties but also by geographical factors. Moreover, the identification of its influencing factors is challenging because of the intricate interactive effects among them. Previous studies primarily focused on single-factor identification and spatial distribution characterization, neglecting the characteristics and spatial features of soil heavy metal concentration under the interactive effects of geographical factors and soil properties. This study assessed the influence of geographical factors, soil properties, and their interactive effects on the spatial distribution of soil arsenic (As), in a typical arsenic-contaminated paddy field area by employing machine learning, analysis of variance, and spatial analysis methods. The findings show that the prediction performance (R2) of the random forest model for soil As concentration was 0.596, and the primary factors influencing the distribution of soil As are elevation, roads, rivers, soil pH, and cation exchange capacity (CEC). Moreover, the interactive effect between elevation and soil CEC had a significant effect on soil As (p < 0.05), exhibiting spatially homogeneous characteristics. The interactive effect between rivers and both soil pH and soil CEC exhibited spatially heterogeneous effects on soil As (p < 0.1). Additionally, the interactive effect between roads and soil pH affected soil As (p < 0.05), with spatially homogeneous characteristics. By identifying the main influencing factors of As in paddy soil, this study further explores the variation characteristics of soil As concentration under the interactive effects of geographical factors and soil properties. These insights can serve as a valuable reference for the precise prevention of As pollution in paddy field area.

Prediction of Primary Admission Total Charges Following Single-Level Lumbar Arthrodesis Utilizing Machine Learning.

Study DesignRetrospective analysis utilizing machine learning.ObjectivesThis study aims to identify the key factors influencing total charges during the primary admission period following single-level lumbar arthrodesis, using machine learning models to enhance predictive accuracy.MethodsData were extracted from the National Inpatient Sample (NIS) database and analyzed using various machine learning models, including random forest, gradient boosting trees, and logistic regression. A total of 78,022 unweighted cases of patients who underwent single-level lumbar arthrodesis were identified using the NIS database from 2016 to 2020. Variables included hospital size, region, patient-specific factors, and procedural details. Multivariate linear regression was also used to identify charge-related variables.ResultsThe average total charge for single-level lumbar arthrodesis was $145,600 ± $102,500. Significant predictors of charge included length of stay, hospital size, hospital ownership, and region. Private investor-owned hospitals and procedures performed in the Western U.S. were associated with higher charges. Random forest models demonstrated superior predictive accuracy with an AUC of .866, outperforming other models.ConclusionsHospital characteristics, regional factors, and patient-specific variables significantly influence the charges of single-level lumbar arthrodesis. Machine learning models, particularly random forest, provide robust tools for predicting healthcare costs, enabling better resource allocation and decision-making. Future research should explore these dynamics further to optimize cost management and improve care quality.

Neutrophil percentage-to-Albumin Ratio as a predictor of acute Kidney Injury in cirrhotic patients:A novel approach utilizing artificial intelligence.

Acute kidney injury (AKI) is common in cirrhotic patients and is associated with a poor prognosis. The neutrophil percentage-to-albumin ratio (NAR) is a novel marker of systemic inflammation. This study aims to evaluate the predictive power of NAR for AKI in cirrhotic patients with ascites using AI tools such as Shapley Additive Explanations and a Random Forest (RF) model. This study involved 322 patients with liver cirrhosis and ascites. AKI was defined based on the International Club of Ascites. The sensitivity and specificity of the NAR were evaluated using receiver operating characteristic curve analysis. Predictors of AKI were identified, and the significance of artificial intelligence features was determined using Shapley Additive Explanations. The RF model was utilized to rank all predictors for AKI. In our study, 130 patients (40.3%) developed AKI, with Stages 2 and 3 analyzed for clinical relevance. The NAR at a cutoff of >23.2 revealed a substantial predictive value for AKI stages 2 and 3 (AUC = 0.893, sensitivity = 85.4%, specificity = 72.3%).The RF model identified serum albumin as the top predictor with an importance score of 0.350, followed by NAR with a score of 0.280, and CKD ranked third with a score of 0.170. These factors are key drivers in predicting the outcome of AKI. Age, ischemic heart disease, diabetes mellitus, INR, and hemoglobin had lower predictive power. NAR could be a new, affordable, and non-invasive test demonstrating a strong discriminative ability to predict AKI in cirrhotic patients.

Acoustic and Natural Language Markers for Bipolar Disorder: A Pilot, mHealth Cross-Sectional Study.

Monitoring symptoms of bipolar disorder (BD) is a challenge faced by mental health services. Speech patterns are crucial in assessing the current experiences, emotions, and thought patterns of people with BD. Natural language processing (NLP) and acoustic signal processing may support ongoing BD assessment within a mobile health (mHealth) framework. Using both acoustic and NLP-based features from the speech of people with BD, we built an app-based tool and tested its feasibility and performance to remotely assess the individual clinical status. We carried out a pilot, observational study, sampling adults diagnosed with BD from the caseload of the Nord Milano Mental Health Trust (Italy) to explore the relationship between selected speech features and symptom severity and to test their potential to remotely assess mental health status. Symptom severity assessment was based on clinician ratings, using the Young Mania Rating Scale (YMRS) and Montgomery-Åsberg Depression Rating Scale (MADRS) for manic and depressive symptoms, respectively. Leveraging a digital health tool embedded in a mobile app, which records and processes speech, participants self-administered verbal performance tasks. Both NLP-based and acoustic features were extracted, testing associations with mood states and exploiting machine learning approaches based on random forest models. We included 32 subjects (mean [SD] age 49.6 [14.3] years; 50% [16/32] females) with a MADRS median (IQR) score of 13 (21) and a YMRS median (IQR) score of 5 (16). Participants freely managed the digital environment of the app, without perceiving it as intrusive and reporting an acceptable system usability level (average score 73.5, SD 19.7). Small-to-moderate correlations between speech features and symptom severity were uncovered, with sex-based differences in predictive capability. Higher latency time (ρ=0.152), increased silences (ρ=0.416), and vocal perturbations correlated with depressive symptomatology. Pressure of speech based on the mean intraword time (ρ=-0.343) and lower voice instability based on jitter-related parameters (ρ ranging from -0.19 to -0.27) were detected for manic symptoms. However, a higher contribution of NLP-based and conversational features, rather than acoustic features, was uncovered, especially for predictive models for depressive symptom severity (NLP-based: R2=0.25, mean squared error [MSE]=110.07, mean absolute error [MAE]=8.17; acoustics: R2=0.11, MSE=133.75, MAE=8.86; combined: R2=0.16; MSE=118.53, MAE=8.68). Remotely collected speech patterns, including both linguistic and acoustic features, are associated with symptom severity levels and may help differentiate clinical conditions in individuals with BD during their mood state assessments. In the future, multimodal, smartphone-integrated digital ecological momentary assessments could serve as a powerful tool for clinical purposes, remotely complementing standard, in-person mental health evaluations.

IoT device for detecting abnormal vibrations in motors using TinyML

This paper presents an innovative approach to motor bearing fault detection using TinyML on an IoT device. We developed a system that integrates spectral analysis and deep learning on a resource-constrained edge device, enabling real-time monitoring and anomaly detection. Our method achieves 96.5(% accuracy in laboratory outperforming baseline Random Forest and SVM models. The system's low latency (300 ms from data collection to alert generation) and computational efficiency make it suitable for real-time industrial applications. We address challenges such as environmental noise and connectivity issues and discuss future directions including multi-modal sensor integration and federated learning. This research contributes to the growing field of edge AI for predictive maintenance, demonstrating the viability of sophisticated machine learning models on low-power microcontrollers.

Improving air quality prediction using hybrid BPSO with BWAO for feature selection and hyperparameters optimization

Air pollution poses a significant threat to public health and environmental sustainability, necessitating accurate predictive models for effective air quality management. This study uses machine learning techniques to forecast air quality through utilizing the annual AQI dataset obtained from the U.S. Environmental Protection Agency (EPA). Feature selection (FS) was conducted using Binary version of Grey Wolf Optimizer (BGWO), Particle Swarm Optimization (BPSO), Whale Optimization Algorithm (BWAO), and a novel hybrid BPSO-BWAO approach to identify the most relevant features for AQI prediction. Among the feature selection methods, BPSO achieved the best Mean Squared Error (MSE) score of 53.56, but with high variance, while BWAO demonstrated lower variance and consistent results. The hybrid BPSO-BWAO method emerged as the optimal solution, achieving an MSE of 53.93 with improved stability and feature set balance, selecting key features such as ‘Days with AQI,’ ‘Median AQI,’ ‘Days CO,’ ‘Days NO2,’ ‘Days PM2.5,’ ‘Good_Days_Percent,’ and ‘Unhealthy_Days_Percent.’ Machine learning models, including Random Forest (RF), Gradient Boosting (GB), K-Nearest Neighbors (KNN), Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), and Linear Regression (LR), were evaluated before and after feature selection. The Random Forest model achieved the best performance after feature selection with an MSE of 53.93, R² of 0.9710, and reduced fitted time. Further optimization using novel hybrid PSO-WAO enhanced RF performance, achieving an improved MSE of 51.82 and R² of 0.9821, demonstrating the efficacy of hyperparameter tuning. The study concludes that feature selection and hyperparameter optimization significantly improve model accuracy and computational efficiency, offering a robust framework for air quality forecasting.

Random Forest model for precise cooling load estimation in optimized and non-optimized form

Abstract Energy is vital for life and human development, with global warming due to activities such as the combustion of fossil fuels and deforestation emitting dangerous greenhouse gases, changing the climate of the Earth. Global energy demand is increasing, with developed nations viewing buildings as major energy consumers. Due to the long lifespan of buildings, it is important to evaluate their suitability to future climate change and possible changes in energy consumption. Appraisal of the cooling loads in each building is now required due to rising energy costs and the need to reduce the impacts of climate change caused by energy consumption from fossil fuels in buildings. This paper aims to apply Random Forest Regression (RF) and Support Vector Regression (SVR), well-known machine learning algorithms to predict building cooling loads. It utilizes the Jellyfish Search Optimizer (JSO) and Transit Search Optimization Algorithm (TSOA) to enhance accuracy and minimize overall error in Cooling Load (CL) estimation. The investigation suggests two high-performance schemes, applies two optimizers for hybrid schemes, and utilizes an ensemble approach for accurate appraisal. Moreover, the SHAP method is utilized to compare the effectiveness of the parameters. The research proves to be insightful in constructing CL projection and suggests that a RFJS-based model is the most effective way to optimize energy consumption. The hybrid model attained an R 2 of 0.994 at its best and RMSE of 0.744. Other than this, the following effective ensemble approach was RSJS, whose R 2 and RMSE were 0.989 and 0.985, accordingly. The third best-performing model was SVJS with R 2 and RMSE values of 0.972 and 1.583, accordingly.

Screening colorectal cancer associated autoantigens through multi-omics analysis and diagnostic performance evaluation of corresponding autoantibodies

BackgroundThis study aims to screen, validate novel biomarkers and develop a user-friendly online tool for the detection of colorectal cancer (CRC).MethodsMulti-omics approach, comprising proteomic analysis and single-cell transcriptomic analysis, was utilized to discover candidate tumor-associated antigens (TAAs). The presence of tumor-associated autoantibodies (TAAbs) in serum was subsequently assessed using enzyme-linked immunosorbent assays (ELISA) in 300 CRC patients and 300 healthy controls. Ten machine learning algorithms were utilized to develop diagnostic models, with the optimal one selected and integrated into an R Shiny-based GUI to enhance usability and accessibility.ResultsWe identified twelve potential TAAs: HMGA1, NPM1, EIF1AX, CKS1B, HSP90AB1, ACTG1, S100A11, maspin, ANXA3, eEF2, P4HB, and HKDC1. ELISA results showed that five TAAbs including anti-CKS1B, anti-S100A11, anti-maspin, anti-ANXA3, and anti-eEF2 were potential diagnostic biomarkers during the diagnostic evaluation phase (all P < 0.05). The Random Forest model yielded an AUC of 0.82 (95% CI: 0.78–0.88) on the training set and 0.75 (95% CI: 0.68–0.82) on the test set, demonstrating the robustness of the results. Web-based implementations of CRC diagnostic tools are publicly accessible via weblink https://qzan.shinyapps.io/CRCPred/.ConclusionsA five biomarker panel can server as complementary biomarker to CEA and CA19-9 in CRC detection.

Machine learning approaches for assessing medication transfer to human breast milk.

The human milk/plasma (M/P) drug concentration ratio is crucial in pharmacology, especially for breastfeeding mothers undergoing treatment. It determines the extent to which drugs ingested by the mother pass into breast milk, potentially affecting the infant. This study conducted a comprehensive evaluation of multiple machine learning algorithms to assess their effectiveness in predicting the M/P ratio. The dataset consists of 162 drugs and 11 predictor variables. M/P ratios were categorized into two groups of (0, 1) and (≥ 1), and a refined three-category system: (0, < 0.5), (0.5, < 1), and (≥ 1). The ML techniques utilized include K-Nearest Neighbors (KNN), Random Forest, Support Vector Machine (SVM), and Neural Networks. We implied the five-fold cross-validation to ensure the model's robustness and Principal Component Analysis (PCA) was applied for data visualization. Bayesian Information Criterion (BIC) was used in the KNN model selection to balance complexity and explanatory power. In our study, KNN achieved average accuracies of 79% for the two-category system and 60% for the three-category. Random Forest models show 77 and 64% average accuracy, respectively. SVM achieved similar results with 78 and 67%, while Neural Networks have the overall best result among the other models with average accuracies of 82 and 76% accuracy. The study highlights the potential of machine learning (ML) techniques in predicting M/P ratios, offering valuable insights for risk assessment during drug development. These predictive models can serve as a valuable tool for estimating drug transfer into breast milk, helping to bridge knowledge gaps in drug safety for lactating individuals. Further validation and refinement by incorporating larger datasets can enhance their reliability and applicability. Advancing these techniques can support safer medication use and informed clinical decision-making for lactating individuals.

Predictive modeling of rapid glaucoma progression based on systemic data from electronic medical records

This study investigated the baseline systemic features that predict rapid thinning of the retinal nerve fiber layer (RNFL) in patients with primary open-angle glaucoma (POAG). A database drawn from electronic medical records (EMRs) was searched for patients diagnosed with POAG between 2009 and 2016 who had been followed up for > 5 years with the annual evaluation of global RNFL thickness using spectral-domain optical coherence tomography. The rate of change in global RNFL thickness for each eye was determined by linear regression analysis over time. Systemic data obtained within 6 months from the time of glaucoma diagnosis were extracted from the EMRs and incorporated into a model to predict the rate of progressive RNFL thinning. The predictive model was trained and tested using a random forest (RF) method and interpreted using Shapley additive explanation plots (SHAP). The features able to explain the rate of progressive RNFL thinning were identified and interpreted. Data from 1256 eyes of 696 patients and 1107 eyes of 607 patients were included in the training and test sets, respectively. The R2 value for the RF model was 0.88 and mean absolute error of the model was 0.205 μm/year. The prediction model identified higher serum levels of aspartate aminotransferase, lower blood glucose, lower systolic blood pressure, and higher high-density lipoprotein as the four most important systemic features predicting rapid RNFL thinning over 5 years. Among the ophthalmic features, a higher global RNFL thickness and a higher intraocular pressure were the most important factors predicting rapid RNFL thinning. The study revealed baseline systemic features from the EMRs that were of predictive value for progression rate of POAG patients.

Habitat suitability modeling to improve conservation strategy of two highly-grazed endemic plant species in saint Catherine Protectorate, Egypt

BackgroundBiodiversity is seriously threatened by climate change impacts in the long term. Conservationists must possess a comprehensive knowledge about habitat suitability of different species and factors that control their distribution in order to effectively minimize biodiversity loss.ResultsThe present study showed the response of two endemic taxa in Saint Catherine protectorate (SKP) (Micromeria serbaliana and Bufonia multiceps) to anticipate climate change over the next few decades using species distribution models. In our analysis, we included the incorporation of bioclimatic variables into the SDM modeling process using four main algorithms: generalized linear model (GLM), Random Forest (RF), Boosted Regression Trees (BRT), and Support Vector Machines (SVM) in an ensemble model. The RF model outperformed other models when analyzing Micromeria serbaliana, whereas BRT demonstrated superiority in the case of Bufonia multiceps. The ensemble models exhibited the best performance, achieving a mean TSS of 0.94 for Micromeria serbaliana and 0.86 for Bufonia multiceps. Micromeria serbaliana was mainly affected by Mean temperature of wettest quarter (Bio8), elevation, and Aridity index. On the other hand, the most significant factors influencing Bufonia multiceps were determined to be Isothermality (Bio2/Bio7) × 100 (Bio3), and elevation. The habitat suitability of Micromeria serbaliana was slightly expanded during the period form 2041–2060, then declined again from 2061 to 2080, while it showed moderate expansion in the case Bufonia multiceps under the two periods.ConclusionThe results of our research support the urgent need for conservation efforts, including reintroduction and planning for in situ and ex situ conservation in appropriate habitats.Clinical trial numberNot applicable.

PREDICTING EMPLOYEE ATTRITION USING TABNET

High employee turnover may threaten stability and productive working environment in a company. Not to mention, it is also more costly than retaining existing employees. The key to solve this problem is to predict employee attrition. Most of the previous researches utilized tree-based model such as Random Forest or a simple deep learning model such as Multi-layer Perceptron. This project will include training a TabNet model for the prediction of employee attrition and comparison of its performance concerning metrics such as accuracy, precision, recall, and F1 score against both a Multi-Layer Perceptron model and a Random Forest model. This study anticipated that the TabNet model would produce results comparable to other models; however, TabNet demonstrated lower performance than both the Random Forest and Multi-Layer Perceptron models. Out of all the models, the Random Forest model performs the best in all key metrics, followed closely by the Multi-Layer Perceptron model. The results indicate that the tree- based algorithms seem to be producing better outputs for predicting employee attrition in structured datasets. The findings of this research offer valuable insights for businesses aiming to improve employee retention strategies.

Approximation of Oxygen Transfer Efficiency of Solid Jet Aerator Having Circular Opening with Kernel Function-Based Models and Random Forest Models

Approximation of Oxygen Transfer Efficiency of Solid Jet Aerator Having Circular Opening with Kernel Function-Based Models and Random Forest Models

Radiomics and machine learning for predicting metachronous liver metastasis in rectal cancer.

A recent study by Long et al used a predictive model to explore the efficacy of radiomics based on multiparametric magnetic resonance imaging in predicting metachronous liver metastasis (MLM) in newly diagnosed rectal cancer (RC) patients. The machine learning algorithms, particularly the random forest model (RFM), appeared well-matched to the complex nature of radiomics data. The predictive capabilities of the RFM, as evidenced by the area under the curve of 0.919 in the training cohort and 0.901 in the validation cohort, highlighted its potential clinical utility. However, we highlighted several methodological limitations, including excluding genomic markers, potential biases from the retrospective design, limited generalizability due to a single-center study, and variability in image interpretation. We propose further investigation into integrating multi-omic data, conducting larger multicenter studies, and utilizing advanced imaging techniques. Additionally, we highlighted the importance of interdisciplinary collaboration to improve predictive model development and advocate for cost-effectiveness analyses to facilitate clinical integration. Overall, this predictive model may improve the early detection and management of MLM in RC patients, with promising avenues for future exploration. Ongoing research in this domain can potentially improve clinical outcomes and the quality of care for RC patients.

Urease in acetogenic Lachnospiraceae drives urea carbon salvage in SCFA pools

ABSTRACT The gut microbiota produces short-chain fatty acids (SCFA) and acidifies the proximal colon which inhibits enteric pathogens. However, for many microbiota constituents, how they themselves resist these stresses is unknown. The anaerobic Lachnospiraceae family, which includes the acetogenic genus Blautia, produce SCFA, are genomically diverse, and vary in their capacity to acidify culture media. Here, we investigated how Lachnospiraceae tolerate pH stress and found that subunits of urease were associated with acidification in a random forest model. Urease cleaves urea into ammonia and carbon dioxide, however the role of urease in the physiology of Lachnospiraceae is unknown. We demonstrate that urease-encoding Blautia show urea-dependent changes in SCFA production, acidification, growth, and, strikingly, urease encoding Blautia directly incorporate the carbon from urea into SCFAs. In contrast, ureolytic Klebsiella pneumoniae or Proteus mirabilis do not show the same urea-dependency or carbon salvage. In agreement, the combination of urease and acetogenesis functions is rare in gut taxa. We find that Lachnospiraceae urease and acetogenesis genes can be co-expressed in healthy individuals and colonization of mice with a ureolytic Blautia reduces urea availability in colon contents demonstrating Blautia urease activity in vivo. In human and mouse microbial communities, the acetogenic recycling of urea carbon into acetate by Blautia leads to the incorporation of urea carbon into butyrate indicating carbon salvage into broader metabolite pools. Altogether, this shows that urea plays a central role in the physiology of health-associated Lachnospiraceae which use urea in a distinct manner that is different from that of ureolytic pathogens.

Social–Ecological Factors and Ecosystem Service Trade-Offs/Synergies in Vegetation Change Zones of Qilian Mountain National Park During 2000–2020

An ecological restoration assessment aims to evaluate whether ecological restoration projects (ERPs) have achieved predefined ecological objectives, such as improving fractional vegetation cover (FVC) and enhancing ecosystem services (ESs), as well as to optimize restoration strategies based on assessment outcomes. Despite recent advancements, current studies still fall short of fully capturing the trade-offs among ESs and identifying the underlying drivers of different vegetation trends. To address these challenges, we applied the Theil–Sen method to delineate vegetation change zones in the Qilian Mountain National Park (QLMNP) between 2000 and 2020, employed bivariate Moran’s I statistics to analyze the trade-offs and synergies among four ESs within these zones, including carbon sequestration (CS), soil conservation (SC), water conservation (WC), and biodiversity maintenance (BIO), and utilized a spatial random forest (SRF) model to explore the main socio-ecological driving factors of vegetation trends and their spatial distribution. Our results revealed significant vegetation recovery in the QLMNP between 2000 and 2020, particularly in regions with initially low FVC. Positive trends in the CS, SC, and BIO highlighted the success of restoration efforts, primarily driven by land conversion to forests and increased precipitation. However, 8.82% of the QLMNP exhibited stagnation or degradation due to rising temperatures and overgrazing, leading to declines in the SC and BIO. Notably, vegetation restoration introduced trade-offs among the ESs, especially in the high FVC areas, where a strong trade-off emerged between FVC and WC. These findings highlight the need for refining restoration strategies to balance water resource allocation. Finally, we integrated vegetation trends, ES relationships, and driving factors to propose grid-based zonal governance plans for the QLMNP, prioritizing WC and FVC enhancement as critical components of future ecological planning. This study serves as a foundation for optimizing restoration strategies in the QLMNP, maintaining and enhancing ESs, while offering actionable insights for fine-grained restoration evaluation and sustainable development planning in other regions.

Current and future habitat suitability of northern fur seals and overlap with the commercial walleye pollock fishery in the eastern Bering Sea

BackgroundUnderstanding the abiotic and biotic drivers of species distribution is critical for climate-informed ecosystem management. We aimed to understand habitat selection of northern fur seals in the eastern Bering Sea, a declining population that is also a key predator of walleye pollock, the target species for the largest U.S. commercial fishery.MethodsWe developed species distribution models using random forest models by combining satellite telemetry data from lactating female fur seals tagged at different rookery complexes on the Pribilof Islands in the eastern Bering Sea with regional ocean model simulations. We explored how data aggregation at two spatial scales (Pribilof-wide and complex-specific) impacted model performance and predicted distributions. Spatial predictions under hindcasted (1992–2018) and projected (2050–2059) physical and biological conditions were used to identify areas of core habitat, overlap with commercial fishery catches, and potential changes in future habitat suitability.ResultsThe most important environmental predictor variables across all models were bathymetry, bottom temperature, and surface temperature. The Pribilof-wide model both under- and overrepresented the importance of specific areas, while complex-specific models exhibited considerable variability in transferability performance. The majority of core habitat occurred on the continental shelf in areas that overlapped with commercial catches of walleye pollock during the “B” season (June – October), with an average of 76% of the total percentage of the catch occurring in core fur seal habitat within the foraging range of lactating females. Projections revealed that considerable changes in fur seal habitat suitability may occur in the coming decades, with complex-specific variation in the magnitude and direction of changes.ConclusionsOur results illustrate the need to sample multiple sites whenever possible and consider spatial scale when extrapolating species distribution model output for central-place foragers, even when terrestrial sites are < 10 km apart. The high overlap between suitable fur seal habitat and commercial fishery catches of pollock, coupled with projected changes in habitat suitability, underscore the need for targeted studies investigating fisheries impacts on this declining population.

Cultivar Differentiation and Origin Tracing of Panax quinquefolius Using Machine Learning Model-DrivenComparative Metabolomics

American ginseng (Panax quinquefolius L.) is a rare and valuable plant utilized for medicinal and culinary purposes, with its geographic origin and cultivation significantly affecting its quality and efficacy. However, the metabolic differences between cultivated and wild American ginseng are not well understood. An accurate and reliable method for tracing the origin and evaluating the quality of American ginseng is therefore urgently required. This study introduces a UHPLC-Q/TOF-MS-based comparative metabolomics and machine learning strategy for the rapid identification of wild and cultivated American ginseng. Both principal component analysis and hierarchical cluster analysis revealed distinct metabolic phenotypes between wild and cultivated American ginseng. Furthermore, the integration of univariate and multivariate statistical analyses identified eight differential metabolites in the ESI+ mode and three in the ESI- mode, including seven ginsenosides. A potential ginsenosides marker panel was used to construct five machine learning models to assist in diagnosing the metabolic phenotypes of American ginseng. The Random Forest model, based on the eight differential metabolites in the ESI+ mode, achieved a 100% classification rate in both test and validation sets for distinguishing between wild and cultivated American ginseng. This study highlights the feasibility and application of our artificial intelligence-driven comparative metabolomics strategy for cultivar identification and geographic tracing of American ginseng, offering new insights into the molecular basis of metabolic variation in cultivated American ginseng.

Analysis of risk factors of social frailty in older adults living with HIV/AIDS

As a predictor of mortality, accidental disability, depressive symptoms, and other adverse outcomes in the older population, social frailty has yet to receive attention among older adults living with HIV/AIDS. Therefore, this study, using Least Absolute Shrinkage and Selection Operator (LASSO) and Random Forest (RF), aims to analyze the current status and risk factors of social frailty in older adults living with HIV/AIDS and provides a basis for its prevention and reversal. From January to December 2024, a total of 335 older adults living with HIV/AIDS (≥ 50 years old) receiving outpatient treatment at a tertiary hospital in Zunyi, Guizhou, China, were selected using convenience sampling. Data were collected through a general information questionnaire (demographic and clinical disease), the HALFT scale for social frailty, the Geriatric Depression Scale-15 (GDS-15), the Barthel Index (BI), the Montreal Cognitive Assessment (MoCA), the Pittsburgh Sleep Quality Index (PSQI), and the Social Support Rating Scale (SSRS). Data processing and analysis were conducted using SPSS 29.0, Python 3.11, and R 4.3.1. Among the 335 older adults living with HIV/AIDS, 105 cases (31.34%) exhibited social frailty. LASSO regression was used for validation, and based on the lambda.min value (λ = 0.0127), 12 relevant variables with non-zero coefficients were selected from 26 variables. These 12 variables were then incorporated into the RF model, which identified that older adults living with HIV/AIDS with reduced activities of daily living (ADL), lack of exercise, cognitive impairment, depression, lower social support, older age, sleep disorders, low CD4+ lymphocyte counts, low per capita household income, complications, smoking history, and those enrolled in China’s free antiretroviral therapy (ART) program faced a higher risk of social frailty. The top five risk factors in order of importance were ADL, exercise, cognitive impairment, depression, and social support. The incidence of social frailty is relatively high among older adults living with HIV/AIDS. The risk factors include ADL, exercise, cognitive impairment, depression, social support, age, sleep disorders, CD4+ lymphocyte count, per capita household income, complications, smoking history, and participation in China’s free ART program. This highlights the importance of healthcare institutions and community healthcare workers addressing social frailty in older adults living with HIV/AIDS. Early identification of the risk factors can help prevent or reverse the onset and progression of social frailty, improving prognosis and quality of life. It also provides a theoretical foundation for developing targeted intervention measures.