Search Results (628)

We present LazyNet, a compact one-step neural-ODE model for single-cell CRISPR activation/interference (A/I) that operates directly on two-snapshot (“pre → post”) measurements and yields parameters with clear mechanistic meaning. The core log–linear–exp residual block exactly represents multiplicative effects, so synergistic multi-locus responses appear as explicit components rather than opaque composites. On a 53k-cell × 18k-gene neuronal Perturb-seq matrix, a three-replica LazyNet ensemble trained under a matched 1 h budget achieved strong threshold-free ranking and competitive error (genome-wide r ≈ 0.67) while running on CPUs. For comparison, we instantiated transformer (scGPT-style) and state-space (RetNet/CellFM-style) architectures from random initialization and trained them from scratch on the same dataset and within the same 1 h cap on a GPU platform, without any large-scale pretraining or external data. Under these strictly controlled, low-data conditions, LazyNet matched or exceeded their predictive performance while using far fewer parameters and resources. A T-cell screen included only for generalization showed the same ranking advantage under the identical evaluation pipeline. Beyond prediction, LazyNet exposes directed, local elasticities; averaging Jacobians across replicas produces a consensus interaction matrix from which compact subgraphs are extracted and evaluated at the module level. The resulting networks show coherent enrichment against authoritative resources (large-scale co-expression and curated functional associations) and concordance with orthogonal GPX4-knockout proteomes, recovering known ferroptosis regulators and nominating testable links in a lysosomal–mitochondrial–immune module. These results position LazyNet as a practical option for from-scratch, low-data CRISPR A/I studies where large-scale pretraining of foundation models is not feasible. Full article

Routine milk-recording data may provide valuable insights into dairy cow welfare, although their ability to accurately reflect herd-level welfare outcomes remains unclear. This study explored the associations between routinely collected milk biomarkers and farm-level welfare status using a comparative machine learning approach. Using the Welfare Quality^® (WQ^®) protocol, 43 commercial dairy farms were classified as Enhanced, Acceptable, or Not Classified. Farm-level milk variables included somatic cell count (SCC), differential somatic cell count (DSCC), fat-to-protein ratio (FPR), fat, protein, casein, lactose, urea, β-hydroxybutyrate (BHB), acetone, total plate count (TPC), and morning milk yield. Kruskal–Wallis tests revealed significant differences among welfare classes for DSCC, SCC, lactose, and milk yield (False Discovery Rate-adjusted p < 0.05). Six machine learning algorithms were trained using 10-fold stratified cross-validation. The Elastic-Net (ENET) model showed the highest mean performance (Accuracy = 0.72 ± 0.19; Kappa = 0.56 ± 0.31), followed by Random Forest and Multilayer Perceptron (Accuracy = 0.70). Model accuracy exhibited substantial variability across cross-validation folds, reflecting the limited sample size and class imbalance. Across models, the most influential variables were SCC, DSCC, lactose, milk yield, FPR, fat, and urea. Overall, the findings provide preliminary and exploratory evidence that routine milk biomarkers capture welfare-relevant patterns at the herd level, supporting their potential role as complementary indicators within data-driven welfare assessment frameworks. Full article

Background: The clinical–radiological paradox in multiple sclerosis (MS) underscores the need for biomarkers that better reflect neurodegenerative pathology. Serum neurofilament light chain (sNfL) is a dynamic marker of neuroaxonal injury, while brain volumetry provides structural assessment of disease impact. However, the precise link between sNfL and regional atrophy patterns, as well as their combined utility for patient stratification and prediction, remains underexplored. Objective: This study aimed to establish a multimodal biomarker framework by integrating sNfL with comprehensive volumetric MRI to define neurodegenerative endophenotypes and predict neuroaxonal injury using Bayesian inference and machine learning. Methods: In a cohort of 57 MS patients, sNfL levels were measured using single-molecule array (Simoa) technology. Brain volumes for 42 regions were quantified via automated deep learning segmentation (mdbrain software). We employed (1) Bayesian correlation to quantify evidence for sNfL–volumetric associations; (2) mediation analysis to test whether grey matter atrophy mediates the EDSS–sNfL (Expanded Disability Status Scale) relationship; (3) unsupervised K-means clustering to identify patient subtypes based on combined sNfL–volumetric profiles; and (4) supervised machine learning (Elastic Net and Random Forest regression) to predict sNfL from volumetric features. Results: Bayesian analysis revealed strong evidence linking sNfL to total grey matter volume (r = −0.449, BF₁₀ = 0.022) and lateral ventricular volume (r = 0.349, BF₁₀ = 0.285). Mediation confirmed that grey matter atrophy significantly mediates the relationship between EDSS and sNfL (indirect effect = 0.45, 95% CI [0.20, 0.75]). Unsupervised clustering identified three distinct endophenotypes: “High Neurodegeneration” (elevated sNfL, severe atrophy, high disability), “Moderate Injury,” and “Benign Volumetry” (low sNfL, preserved volumes, mild disability). Supervised models predicted sNfL with high accuracy (R² = 0.65), identifying total grey matter volume, ventricular volume, and age as top predictors. Conclusions: This integrative multi-method analysis demonstrates that sNfL is robustly associated with global grey matter and ventricular volumes, and that these measures define clinically meaningful neurodegenerative subtypes in MS. Machine learning confirms that a concise set of volumetric features can effectively predict neuroaxonal injury. These findings advance a pathobiology-driven subtyping framework and provide a validated model for using routine MRI volumetry to assess neuroaxonal health, with implications for prognosis and personalised therapeutic strategies. Full article

Amid accelerating climate change, intensifying urban heat island effects, and rising public demand for livable, walkable streets, there is an urgent practical need for interpretable and actionable evidence on streetscape quality. Yet, research on streetscape quality has often relied on single data sources and linear models, limiting insight into multidimensional perception; evidence from temperate monsoon cities remains scarce. Using Tianjin’s main urban area as a case study, we integrate street-view imagery with remote sensing imagery to characterize satellite-derived environmental indicators at the point scale and examine the following five perceptual outcomes: comfort, aesthetics, perceived greenness, summer heat perception, and willingness to linger. We develop a three-step interpretable assessment, as follows: Elastic Net logistic regression to establish directional and magnitude baselines; Generalized Additive Models with a logistic link to recover nonlinear patterns and threshold bands with Benjamini–Hochberg false discovery rate control and binned probability calibration; and Shapley additive explanations to provide parallel validation and global and local explanations. The results show that the Green View Index is consistently and positively associated with all five outcomes, whereas Spatial Balance is negative across the observed range. Sky View Factor and the Building Visibility Index display heterogeneous forms, including monotonic, U-shaped, and inverted-U patterns across outcomes; Normalized Difference Vegetation Index and Land Surface Temperature are likewise predominantly nonlinear with peak sensitivity in the midrange. In total, 54 of 55 smoothing terms remain significant after Benjamini–Hochberg false discovery rate correction. The summer heat perception outcome is highly imbalanced: 94.2% of samples are labeled positive. Overall calibration is good. On a standardized scale, we delineate optimal and risk intervals for key indicators and demonstrate the complementary explanatory value of street-view imagery and remote sensing imagery for people-centered perceptions. In Tianjin, a temperate monsoon megacity, the framework provides reproducible, actionable, design-relevant evidence to inform streetscape optimization and offers a template that can be adapted to other cities, subject to local calibration. Full article

This work presents the development of a low-cost and high-performance multi-sensory gas detection instrument named the AIMNet Sensor, with the integration of a machine learning-based data processing method. The compact and low-power instrument (8.5 × 11.5 cm, 1.4 W) houses the core sensing hardware module, Senseair K96, that integrates both a non-dispersive infrared (NDIR)-based gas sensing unit and a BME280 environmental sensing unit. To address the outdoor operation challenges caused by environmental fluctuation due to the varying temperature, humidity, and pressure, from the software aspect, multiple machine learning-based regression models were trained in this work on 13,125 calibration data points collected under controlled laboratory conditions. Among ten tested algorithms, the Multilayer Perceptron (MLP) and Elastic Net models achieved the highest accuracy, with R-squared coefficient $R^2>0.8$ on both indoor and outdoor scenarios, and with inter-sensor root mean square error (RMSE) within 1.5 ppm across four identical instruments. Moreover, field mobile validation was performed near a wastewater management facility using this solution, confirming a strong correlation with LI-COR reference measurements and a reliable detection of ${\mathrm{CH}}_4$ leaks with concentrations up to 18 ppm at the test site. Overall, this machine learning-integrated NDIR sensing solution (i.e., AIMNet) offers a practical and scalable solution towards a more robust distributed ${\mathrm{CH}}_4$ monitoring network for real-world field-deployable applications. Full article

This study quantifies the decarbonisation potential of enhanced material circularity in the EU27 over the 2015–2022 period by integrating material flow data with elasticity-based emissions modelling. Using panel regression and logarithmic mean Divisia index (LMDI) decomposition, we evaluate the influence of recycling rate acceleration and material intensity decline on material-embedded emissions over the 2015–2022 period. The findings indicate that although recycling rates increased by 42% during this time, virgin materials remain responsible for over 97% of emissions. Decomposition results reveal that intensity improvements—measured as a cumulative LMDI intensity effect of −0.867 log-change units, equivalent to approximately a 58% reduction in emissions—offset most of the upward pressure from growing material demand and shifting composition. Scenario projections to 2050, based on empirically derived elasticities, show that accelerated circular economy pathways—assuming 4% annual growth in recycling rates and a 3% decline in material intensity—can reduce emissions by over 90%. In contrast, baseline policies fall short of net-zero targets. Sensitivity analysis confirms that policy ambition dominates parameter uncertainty in shaping future emissions trajectories. The study highlights the critical role of combined demand-side and supply-side measures in aligning material consumption with climate goals. The study highlights the crucial role of combined demand-side and supply-side measures in aligning material consumption with climate goals and advancing progress toward Sustainable Development Goal 12 (Responsible Consumption and Production). Full article

Railroad-highway grade crossings (RHGCs) are critical points of conflict between roadway and rail systems, contributing to over 2000 crashes and 250 fatalities annually in the United States. This study applied machine learning methods (ML) techniques to model and predict crash frequency at RHGCs, using a comprehensive dataset from the Federal Railroad Administration (FRA) and Tennessee Department of Transportation (TDOT). The dataset included 807 validated crossings, incorporating roadway geometry, traffic volumes, rail characteristics, and control features. Five ML models—Random Forest, XGBoost, PSO-Elastic Net, Transformer-CNN, and Autoencoder-MLP—were developed and compared to a traditional Negative Binomial (NB) regression model. Results showed that ML models significantly outperformed the NB model in predictive accuracy, with the Transformer-CNN achieving the lowest Mean Squared Error (21.4) and Mean Absolute Error (3.2). Feature importance analysis using SHAP values consistently identified Annual Average Daily Traffic (AADT), Truck Traffic Percentage, and Number of Lanes as the most influential predictors, findings that were underrepresented or statistically insignificant in the NB model. Notably, the NB model failed to detect the nonlinear relationships and interaction effects that ML algorithms captured effectively. While only three variables were statistically significant in the NB model, ML models revealed a broader spectrum of critical crash determinants, offering deeper interpretability and higher sensitivity. These findings emphasize the superiority of machine learning approaches in modeling RHGC safety and highlight their potential to support data-driven interventions and policy decisions for reducing crash risks at grade crossings. Full article

Aviation accounts for a disproportionate share of tourism-related carbon emissions. Many travelers express environmental concern but continue to fly, reflecting the well-documented attitude–behavior gap. This study examines the concept of flight behavioral integrity (i.e., the alignment between professed avoidance of air travel for environmental reasons and actual flying behavior) to assess whether integrity profiles predict support for climate policy. Drawing on nationally representative survey data from Germany (N = 2410), respondents were classified into four groups based on flight avoidance attitudes and reported flight activity in the past 12 months. An elastic-net multinomial regression tested psychological predictors of group membership, and factorial ANCOVAs assessed differences in environmental and climate policy support. Results showed that flight avoidance attitudes, rather than recent flying behavior, were the primary predictors of both integrity profiles and policy support. Flight-avoidant respondents consistently reported stronger policy endorsement, regardless of whether they had flown. Contrary to expectations, recent fliers expressed marginally higher support than non-fliers, potentially reflecting compensatory mechanisms or sociodemographic factors. Findings suggest that there are opportunities for tourism operators and policymakers to engage travelers through value-based (vs. purely behavioral) sustainability initiatives. Full article

Background/Objectives: This paper presents the results of predicting chronological age from psychophysiological tests using machine learning regressors. Methods: Subjects completed a series of psychological tests measuring various cognitive functions, including reaction time and cognitive conflict, short-term memory, verbal functions, and color and spatial perception. The sample included 99 subjects, 68 percent of whom were men and 32 percent were women. Based on the test results, 43 features were generated. To determine the optimal feature selection method, several approaches were tested alongside the regression models using MAE, $R^2$, and $CV\text{\_}{R}^2$ metrics. SHAP and Permutation Importance (via Random Forest) delivered the best performance with 10 features. Features selected through Permutation Importance were used in subsequent analyses. To predict participants’ age from psychophysiological test results, we evaluated several regression models, including Random Forest, Extra Trees, Gradient Boosting, SVR, Linear Regression, LassoCV, RidgeCV, ElasticNetCV, AdaBoost, and Bagging. Model performance was compared using the determination coefficient ($R^2$) and mean absolute error (MAE). Cross-validated performance ($CV\text{\_}{R}^2$) was estimated via 5-fold cross-validation. To assess metric stability and uncertainty, bootstrapping (1000 resamples) was applied to the test set, yielding distributions of MAE and RMSE from which mean values and 95% confidence intervals were derived. Results: The study identified RidgeCV with winsorization and standardization as the best model for predicting cognitive age, achieving a mean absolute error of 5.7 years and an $R^2$ of 0.60. Feature importance was evaluated using SHAP values and permutation importance. SHAP analysis showed that stroop_time_color and stroop_var_attempt_time were the strongest predictors, followed by several task-timing features with moderate contributions. Permutation importance confirmed this ranking, with these two features causing the largest performance drop when permuted. Partial dependence plots further indicated clear positive relationships between these key features and predicted age. Correlation analysis stratified by sex revealed that most features were significantly associated with age, with stronger effects generally observed in men. Conclusions: Feature selection revealed Stroop timing measures and task-related metrics from math and campimetry tests as the strongest predictors, reflecting core cognitive processes linked to aging. The results underscore the value of careful outlier handling, feature selection, and interpretable regularized models for analyzing psychophysiological data. Future work should include longitudinal studies and integration with biological markers to further improve clinical relevance. Full article

Chlorophyll content is a key physiological indicator reflecting photosynthetic capacity, and the Soil–Plant Analysis Development (SPAD) meter is a commonly used tool for its rapid and non-destructive estimation. Hyperspectral imaging (HSI) is a non-destructive technique that captures fine spectral characteristics and thus holds great potential for high-throughput phenotyping and early stress detection. This study aimed to explore the potential of HSI combined with ensemble learning (EL) to estimate SPAD of rapeseed seedlings under different durations of waterlogging. Hyperspectral images and corresponding SPAD values were collected from six rapeseed cultivars at 0, 2, 4 and 6 days of waterlogging. The mutual information was employed to select the top 30 most relevant spectral and vegetation index features. The EL model was constructed using partial least squares, support vector machine, random forest, ridge regression and elastic net as the first-layer learners and a multiple linear regression as the second-layer learner. The results showed that the EL model showed superior stability and higher prediction accuracy compared to single models across various genotypes and waterlogging treatment datasets. As waterlogging duration increased, the overall model accuracy improved; notably, under 6 days of waterlogging, the EL model achieved an R² of 0.79 and an RMSE of 3.27, indicating strong predictive capability. This study demonstrated that combining EL with HSI enables stable and accurate estimation of SPAD values, therefore providing an effective approach for early stress monitoring in crops. Full article

Three-dimensional cell model systems such as tumour organoids allow for in vitro modelling of self-organized tissue with functional and histologic similarity to in vivo tissue. However, there is a need for standard protocols and techniques to confirm the presence of cancer within organoids derived from tumour tissue. The aim of this study was to assess the utility of a Nanostring gene expression-based machine learning classifier to determine the presence of cancer or normal organoids in cultures developed from both benign and cancerous stomach biopsies. A prospective cohort of normal and cancer stomach biopsies were collected from 2019 to 2022. Tissue specimens were processed for formalin-fixed paraffin-embedding (FFPE) and a subset of specimens were established in organoid cultures. Specimens were labelled as normal or cancer according to analysis of the FFPE tissue by two pathologists. The gene expression in FFPE and organoid tissue was measured using a 107 gene Nanostring codeset and normalized using the Removal of Unwanted Variation III algorithm. Our machine learning model was developed using five-fold nested cross-validation to classify normal or cancer gastric tissue from publicly available Asian Cancer Research Group (ACRG) gene expression data. The models were externally validated using the Cancer Genome Atlas (TCGA), as well as our own FFPE and organoid gene expression data. A total of 60 samples were collected, including 38 cancer FFPE specimens, 5 normal FFPE specimens, 12 cancer organoids, and 5 normal organoids. The optimal model design used a Least Absolute Shrinkage and Selection Operator model for feature selection and an ElasticNet model for classification, yielding area under the curve (AUC) values of 0.99 [95% CI: 0.99–1], 0.90 [95% CI: 0.87–0.93], and 0.79 [95% CI: 0.74–0.84] for ACRG (internal test), FFPE, and organoid (external test) data, respectively. The performance of our final model on external data achieved AUC values of 0.99 [95% CI: 0.98–1], 0.94 [95% CI: 0.86–1], and 0.85 [95% CI: 0.63–1] for TCGA, FFPE, and organoid specimens, respectively. Using a public database to create a machine learning model in combination with a Nanostring gene expression assay allows us to allocate organoids and their paired whole tissue samples. This platform yielded reasonable accuracy for FFPE and organoid specimens, with the former being more accurate. This study re-affirms that although organoids are a high-fidelity model, there are still limitations in validating the recapitulation of cancer in vitro. Full article

Accurate prediction of photosynthetic parameters is pivotal for precision viticulture, as it enables non-invasive monitoring of plant physiological status and informed management decisions. In this study, spectral reflectance data were used to predict key photosynthetic parameters such as assimilation rate (A), effective photosystem II (PSII) quantum yield (${\Phi }_{\mathrm{PSII}}$), and electron transport rate (ETR), as well as stem and leaf water potential (${\Psi }_{\mathrm{stem}}$ and ${\Psi }_{\mathrm{leaf}}$), in Vitis vinifera (cv. Müller-Thurgau) grown in an experimental vineyard in Lower Franconia (Germany). Measurements were obtained on 25 July, 7 August, and 12 August 2024 using a LI-COR LI-6800 system and a PSR+ hyperspectral spectroradiometer. Various machine learning models (SVR, Lasso, ElasticNet, Ridge, PLSR, a simple ANN, and Random Forest) were evaluated, both as standalone predictors and as base learners in a stacking ensemble regressor with a Random Forest meta-learner. First derivative reflectance (FDR) preprocessing enhanced predictive performance, particularly for ${\Phi }_{\mathrm{PSII}}$ and ETR, with the ensemble approach achieving R² values up to 0.92 for ${\Phi }_{\mathrm{PSII}}$ and 0.85 for A at 1 nm resolution. At coarser spectral resolutions, predictive accuracy declined, though FDR preprocessing provided some mitigation of the performance loss. Diurnal patterns revealed that morning to mid-morning measurements, particularly between 9:00 and 11:00, captured peak photosynthetic activity, making them optimal for assessing vine vigor, while midday water potential declines indicated favorable timing for irrigation scheduling. These findings demonstrate the potential of integrating hyperspectral data with ensemble machine learning and FDR preprocessing for accurate, scalable, and high-throughput monitoring of grapevine physiology, supporting real-time vineyard management and the use of cost-effective sensors under diverse environmental conditions. Full article

The accurate prediction of plant height is crucial for optimizing soybean cultivar selection and improving yield estimations. In this study, we investigate the potential of machine learning (ML) algorithms to predict soybean plant height (PH) based on a diverse set of agronomic parameters analyzed from forty soybean cultivars evaluated across sequential harvests. Using a comprehensive dataset, the models Elastic Net (EN), Extra Trees (ET), Gaussian Process Regressor (GPR), K-Nearest Neighbors, and XGBoost (XGB) were compared in terms of predictive accuracy, uncertainty, and robustness. Our results demonstrate that ET outperformed other models with an average correlation coefficient of 0.674, $R^2$ of 0.426 and the lowest RMSE of 6.859 cm and MAE of 5.361 cm, while also showing the lowest uncertainty (5.07%). The proposed ML framework includes an extensive model evaluation pipeline that incorporates the Performance Index (PI), ANOVA, and feature importance analysis, providing a multidimensional perspective on model behavior. The most influential features for PH prediction were the number of stems (NS) and insertion of the first pod (IFP). This research highlights the viability of integrating explainable ML techniques into agricultural decision support systems, enabling data-driven strategies for cultivar evaluation and phenotypic trait forecasting. Full article

Extracellular vesicles (EVs) are increasingly recognized as promising non-invasive biomarkers for cancer and other diseases, but their clinical translation remains limited by the lack of comprehensive characterization strategies. Spectroscopic approaches such as Fourier-transform infrared (FTIR) spectroscopy can provide a global biochemical fingerprint of intact EVs, but their interpretation requires advanced analytical tools. In this study, we applied an autoencoder-based framework to attenuated total reflection FTIR (ATR-FTIR) spectra of blood-derived components, including plasma, red blood cells (RBCs), RBC-ghosts, and EVs, comprising 278 samples collected from 135 patients, to obtain latent features capable of capturing biologically meaningful variability. The autoencoder compressed spectra into 12 latent features while preserving spectral information with low reconstruction error. Unsupervised UMAP projection of the latent features separated the blood components into different clusters, supporting their biological relevance. The model was then applied to EV spectra from patients with hepatocellular carcinoma (HCC) and cirrhotic controls. Four features significantly differed between the two groups, and an elastic-net regularized logistic model evaluated with a leave-one-out cross-validation framework retained a single latent feature, achieving an out-of-fold ROC AUC of 0.785 (95% CI 0.602–0.967), with performance broadly comparable to that typically reported for AFP, the most commonly used biomarker for HCC. This study provides the first proof-of-concept that an autoencoder can be applied to FTIR spectra of EVs, extracting biologically relevant latent features with potential application in cancer detection. Full article

Steroid hormones (SHs), including sex steroids and corticosteroids, are crucial for a healthy pregnancy. We aimed to comprehensively characterize the maternal SH metabolome in late pregnancy and identify clinical, lifestyle, and sociodemographic determinants influencing SH metabolism with a replication in an independent cohort. Urinary SH metabolites were analyzed in 1221 third-trimester pregnant women (aged 28 to 37 years) from two Spanish cohorts, BiSC (2018–2021, n = 721) and INMA-Sabadell (2004–2006, n = 500), using targeted UHPLC-MS/MS. We quantified 50 SH metabolites, resulting in 13 hormone groups, 9 sulfate/glucuronide ratios, and 17 estimated steroid enzymatic activities across steroidogenesis pathways. We applied elastic net regression to identify determinants, and multivariable linear regression models to estimate variance explained. Among the 47 and 28 determinants from BiSC and INMA-Sabadell, respectively, 10 determinant-SH metabolome pairs showed statistically significant associations (p < 0.05), supporting robust replication. Maternal BMI was the main determinant linked to higher corticosteroid and androgen metabolites. Higher physical activity was associated with lower glucocorticoids and progestogen metabolites, while older maternal age was related with lower levels of androgen and corticosteroid metabolites. Tobacco exposure in the first trimester predicted higher levels of cortisol metabolites. Latin American women had lower cortolone levels compared with Spanish women. Parity, dietary fat intake, sleep, alcohol intake, and sex of the fetus contributed to smaller variations in different SHs. This dual-cohort analysis provides the most detailed and replicated evidence to date of how clinical, lifestyle, and sociodemographic factors shape the maternal SH metabolome during late pregnancy. Full article