2,4-Dinitrophenol (2,4-DNP) and p-nitroaniline (PNA), as hazardous nitroaromatic explosives and persistent organic pollutants, exhibit high toxicity, nonbiodegradability, and carcinogenicity. Exposure to these compounds can cause severe health issues, including cardiovascular, neurological, and renal damage. In this study, a novel zirconium-based metal-organic framework (Zr-MOF, denoted as Zr-BPBI) was synthesized via a solvothermal reaction using the ligand 4,4',4″,4‴-([1,1'-biphenyl]-4,4'-diyldi(1H-imidazole-2,4,5-triyl))tetrabenzoic acid (H4BPBI) and ZrCl4. The investigations showed that the original synthetic parameters, including surfactant concentration, acid type, reaction time, and temperature, could strongly affect the morphology, particle size, and fluorescence performance of the as-prepared Zr-BPBI. Interestingly, under excitation at 422 nm, the as-prepared Zr-BPBI nanoparticles (Zr-BPBI-NP) emitted blue luminescence centered at 502 nm, which demonstrated exceptional sensitivity toward PNA and 2,4-DNP with rapid fluorescence quenching responses (within 30 s for 2,4-DNP and 15 s for PNA). Simultaneously, common interfering substances (e.g., small molecules and ions) did not affect the above detection, suggesting high selectivity of the as-prepared Zr-BPBI-NP fluorescent probe. The corresponding limits of detection were 0.36 μM for 2,4-DNP and 0.4 μM for PNA. Obviously, the present Zr-BPBI-NP fluorescent probe is a new option for the highly sensitive and selective detection of PNA and 2,4-DNP in actual aqueous environments.
To evaluate the diagnostic performance of deep learning-based radiomics (DL) and hand-crafted radiomics (HCR) in differentiating benign from malignant orbital tumors. A retrospective analysis was performed on CT data from 145 patients (48 benign, 97 malignant) diagnosed between December 2014 and March 2024. Two radiologists independently assessed conventional CT semantic features (e.g., lesion location, margin definition, internal density homogeneity, calcification, necrosis, and enhancement pattern). Deep transfer learning (DTL) extracted DL features, while traditional methods were used to obtain HCR features. Feature fusion, selection, and modeling were performed using the Least Absolute Shrinkage and Selection Operator (LASSO). Pathological diagnosis served as the gold standard. Model performance was evaluated using receiver operating characteristic (ROC) curves. A nomogram integrating clinical data and significant semantic features was constructed for visualization. The DeLong test and decision curve analysis (DCA) assessed model effectiveness. Multivariate analysis confirmed that homogeneous enhancement and ill-defined/infiltrative margins were independent CT features differentiating benign from malignant tumors. A total of 14 HCR and 30 DL features were extracted; 36 features were retained after fusion. The HCR, DL, fused, and nomogram models achieved AUCs of (0.859/0.816), (0.957/0.826), (0.986/0.811), and (0.975/0.837) in the training and test cohorts, respectively. The DeLong test showed no significant difference between the fused model and the nomogram in either cohort (P = 0.090 and P = 0.198), whereas differences for other model pairs were significant (P < 0.05). DCA indicated that the nomogram provided higher clinical utility. The fused model outperformed single radiomics approaches in accuracy. The nomogram, which integrates clinical data and semantic features, demonstrated superior predictive performance and may support clinical decision-making, particularly for patients who cannot undergo invasive procedures.
Precision-driven nanomaterial approaches continue to show enormous promise and are forecast to transform contemporary medicine. The core strength lies in the ability to engineer materials at the nanoscale to achieve unique physicochemical properties. A theory that is paramount to its unlimited future promise is in creating "designer" or "intelligent" nanomaterials precisely crafted to enable specific biological interactions for applications in health, disease, and infection. Fuelled by their ease of synthesis, catalytic nature, complex surface character, and tunability, the development of next-generation reducible metal oxide nanozymes (rNZs) that mimic the complex perceptive and adaptive capabilities of natural enzymes is a topic of remarkable curiosity. In pursuit of decoding rNZ catalytic mechanisms, this review spotlights the critical contribution of atomistic features. The promise of rNZs to circumvent the therapeutic insufficiencies surrounding bacterial antimicrobial resistance is confirmed to be a viable route forward, and the first evidence of the use of ionizing radiation to augment activity is presented as a novel antibacterial strategy. In some respects, the micromechanisms remain cryptic and elusive. The data reveal the critical contribution of crystal facets and oxygen vacancies within an orchestrated, heterogeneous, tunable, but strikingly complex system. Deciphering these sophisticated behaviors is arguably the next frontier in the field.
Cryo-electron tomography (cryo-ET) enables in situ three-dimensional visualization of many protein complexes and other macromolecular assemblies such as ribosomes in cells, yet automated macromolecule particle identification in 3D cryo-ET tomograms remains a major bottleneck due to dose-limited low signal-to-noise ratios, missing-wedge artifacts, and densely crowded cellular backgrounds. We present TomoSwin3D, an end-to-end three-dimensional (3D) macromolecule particle identification and classification pipeline centered on a Swin Transformer-based U-Net that performs particle identification and classification and outputs particle centroid coordinates. TomoSwin3D leverages a multi-channel input representation that augments raw tomogram densities with complementary 3D feature maps capturing edge strength (Sobel gradients), local contrast enhancement (morphological top-hat), and multiscale blob responses (Difference-of-Gaussians), improving detectability of small and low-contrast targets. To better preserve particle geometry and avoid hand-crafted shape assumptions, it adopts occupancy-preserving supervision that directly uses available 3D instance masks rather than heuristic Gaussian/spherical labels and applies scalable patch-wise inference followed by lightweight post-processing (connected-component analysis, size filtering, centroid extraction) for robust centroid coordinate extraction. Across diverse simulated and experimental cryo-ET tomogram benchmarks including SHREC 2021 and 2020 test datasets, EMPIAR dataset, and Cryo-ET data portal dataset, TomoSwin3D achieves strong and consistent performance in detecting proteins and other particles, outperforming existing methods, with a pronounced advantage in picking hard, small protein particles. These results establish TomoSwin3D as a scalable and accurate solution for high-throughput cryo-ET macromolecule particle picking and downstream subtomogram averaging.
Gene essentiality, the requirement of a gene for survival or proliferation, is central to understanding cellular processes and identifying drug targets. Experimental determination requires large growth screens that are time-consuming and expensive, motivating in silico approaches. Existing methods predominantly use flux balance analysis (FBA), a constraint-based optimisation framework that requires a predefined cellular objective function. This can introduce observer bias, because the objective often reflects the researcher's assumptions rather than the cell's biological goals. Here, we present FluxGAT, a graph neural network (GNN) that predicts gene essentiality from graphical representations of flux sampling data. Flux sampling removes the need for an explicit objective and instead characterises feasible steady-state fluxes. FluxGAT combines this information with metabolic network topology to learn flux-informed node representations and classify reactions as essential or non-essential. We apply FluxGAT to the iCHO2291 genome-scale model of Chinese hamster ovary cells and Mouse1, a generic mouse model with independent essentiality labels. In both systems, FluxGAT improves sensitivity over FBA while maintaining high precision and specificity, and recovers more experimentally essential genes, especially where FBA predicts very few essentials. These results show that flux-informed GNNs can provide more general gene essentiality predictions across mammalian genome-scale models without hand-crafted objective functions.
Artificial intelligence has advanced cancer pathology, but many systems still depend on hand-crafted features, are hard to explain and rely on fragmented workflows. We introduce SPARK (System of Pathology Agents for Research and Knowledge), a foundational agentic artificial intelligence approach that uses language as a universal interface to autonomously generate biologically driven concepts for tumor analysis. SPARK turns biological ideas into analytical tools and works directly with complex pathology data without extra model training. We evaluated SPARK across 18 patient cohorts spanning five cancer types (lung adenocarcinoma, lung squamous cell carcinoma, colorectal cancer, breast cancer and oropharyngeal squamous cell carcinoma) and more than 5,400 patients with available histopathology images and clinical/follow-up information, in both prognostic and predictive settings and on a well characterized spatial biology breast cancer dataset (patient n = 625). We found that SPARK produced clinically and biologically relevant concepts correlated with prognosis, known pathological variables and predictive biomarkers, including patterns of tumor progression and temporal change inferred from static images. A dedicated module allows for human interaction with SPARK. Further prospective validation is needed to evaluate the clinical utility of the tools created by SPARK. All code, parameters and results are openly released to help researchers and clinicians improve diagnostic precision and deepen tumor biology insights.
Humans engage daily in procedural activities such as cooking a recipe or fixing a bike, which can be described as goal-oriented sequences of key-steps following certain ordering constraints. Task graphs mined from videos or textual descriptions have recently gained popularity as a human-readable, holistic representation of procedural activities encoding a partial ordering over key-steps, and have shown promise in supporting downstream video understanding tasks. While previous works generally relied on hand-crafted procedures to extract task graphs from videos, this paper introduces an approach based on gradient-based maximum likelihood optimization of edge weights, which can be used to directly estimate an adjacency matrix and can also be naturally plugged into more complex neural network architectures. We validate the ability of the proposed approach to generate accurate task graphs on the CaptainCook4D and EgoPER datasets. Moreover, we extend our validation analysis to the EgoProceL dataset, which we manually annotate with task graphs as an additional contribution. The three datasets together constitute a new benchmark for task graph learning, where our approach obtains improvements of +14.5%, +10.2% and +13.6% in $F_{1}$ score, respectively, over previous approaches. Thanks to the differentiability of the proposed framework, we also introduce a feature-based approach for predicting task graphs from key-step textual or video embeddings, which exhibits emerging video understanding abilities. Beyond that, task graphs learned with our approach obtain top performance in the Ego-Exo4D procedure understanding benchmark including 5 different downstream tasks, with gains of up to +4.61%, +0.10%, +5.02%, +8.62%, and +15.16% in finding Previous Keysteps, Optional Keysteps, Procedural Mistakes, Missing Keysteps, and Future Keysteps, respectively. We finally show significant enhancements to the challenging task of online mistake detection in procedural egocentric videos, achieving notable gains of +19.8% and +6.4% in the Assembly101-O and EPIC-Tent-O datasets, respectively, compared to the state of the art. The code for replicating the experiments is available at https://github.com/fpv-iplab/Differentiable-Task-Graph-Learning.
High-quality radiotherapy requires accurate dose delivery to target volumes while protecting organs-at-risk (OARs). However, current clinical workflows remain constrained by labor-intensive multidisciplinary collaboration, prolonged planning cycles, and limited scalability. Intelligent automation capable of integrating clinical knowledge and real-world decision patterns is needed to enhance precision and efficiency in radiotherapy planning. This study proposes MARTP, a \textbf{M}ulti-\textbf{A}gent \textbf{R}adiation \textbf{T}herapy \textbf{P}lanning framework driven by large language models (LLMs), to emulate multidisciplinary clinical workflows and enable end-to-end intelligent radiotherapy planning and evaluation.
 Approach: MARTP coordinates five specialized agents to integrate data analysis, weight adjustment, plan optimization, plan evaluation, and report generation into a unified radiotherapy planning workflow. The framework leverages supervised fine-tuning (SFT) on expert weight adjustment demonstrations to improve adaptation to complex clinical cases, incorporates retrieval-augmented generation (RAG) to ground planning decisions in case-specific knowledge, and employs a predefined model-context protocol (MCP) to enable high-precision treatment plan generation. In addition, reinforcement learning (RL) with expert preference data is used to develop an intelligent plan evaluation mechanism.
Results: Statistical analysis reveals that the dosimetric metrics of plans generated by MARTP exhibit no statistically significant differences compared with expert-crafted clinical plans, while the planning efficiency is substantially improved. In addition, the framework demonstrates robust and safe behavior under abnormal input scenarios, maintaining clinically acceptable outputs. The SFT and RL components further enhance the consistency, semantic accuracy, and reliability of model-generated weight adjustments and plan evaluations.
Significance: MARTP demonstrates that LLMs-driven multi-agent systems can effectively replicate multidisciplinary radiotherapy workflows, generating clinically comparable plans with substantially improved efficiency. The framework provides a promising pathway for integrating intelligent automation into radiotherapy practice, supporting more consistent decision-making and scalable treatment planning.
X-ray spectroscopy provides sensitive, element-specific insight into local geometric and electronic structures, but predictive first-principles simulations can be computationally expensive for large and chemically diverse molecular systems. Recent machine-learning approaches have shown promise in accelerating structure-to-spectrum prediction; however, most directly regress discretized spectral intensities and rely on hand-crafted geometric descriptors centered on the absorbing atom. Herein, we introduce a machine learning framework that encodes a detailed, environment-aware representation of the nuclear structure beyond the absorbing site. The model combines these descriptors with a physically motivated, multiscale Gaussian spectral basis whose coefficients are obtained via ridge projection, enforcing smoothness and spectral consistency. To further enhance robustness across chemical and conformational diversity, we employ a multiscale structural similarity loss that couples geometric and spectral resolution. This integrated approach yields accurate and transferable predictions across a wide range of molecular geometries and chemical environments while maintaining physical interpretability. The proposed framework establishes a physically structured and scalable route to machine-learned X-ray spectroscopy.
People complain that they do not know what to say to soften the blow of social rejection. The adoption of language principles (i.e., avoiding apologies while using positive regard, sincere alternatives, and more words) may mitigate the negative consequences of social rejection. Does training help people independently exhibit greater communication skill or do people fail to perform despite "knowing better?" In Study 1, drift diffusion modeling suggested that training helped people "know it when they see it" (i.e., recognize precrafted wording options which better conveyed the language principles). Study 2 found that training aided people's success in independently crafting wording to convey the language principles. The current research highlights that existing social rejector frameworks and previous research on skill acquisition do not capture the experience of nonpunitive rejectors and presents a way that psychological science can address people's concerns about how to soften the blow of social rejection.
Generalist foundation models (GFMs) are renowned for their exceptional capability and flexibility in diverse tasks. In the field of medicine, while GFMs exhibit superior generalizability, specialist models excel in precision because of their domain-specific knowledge. Here we show a cooperative framework, Generalist-Specialist Collaboration (GSCo), that synergistically combines a powerful generalist model with lightweight specialists. In this framework, specialists provide expert guidance, such as diagnostic predictions and visually similar clinical cases, as contextual information to the generalist, which then makes a final diagnosis. We developed MedDr, an open-source GFM tailored for medicine, as well as a suite of lightweight specialist models crafted for specific downstream tasks. A comprehensive evaluation on 32 datasets across diverse medical modalities shows that MedDr outperforms state-of-the-art GFMs on downstream datasets. Furthermore, GSCo exceeds GFMs and specialists in medical image diagnosis and report generation. This approach offers an effective and computationally efficient paradigm for deploying GFMs in clinical settings, enhancing scalability and enabling precise analysis across a wide range of scenarios.
Fluorescence microscopy increasingly produces complex volumetric datasets whose biologically meaningful differences are difficult to capture with hand-crafted measurements, especially when signal is distributed across three-dimensional space. Here, we present an interpretable 3D Bag-of-Visual-Words (BoVW) pipeline for classification and analysis of volumetric microscopy data. The framework detects multiscale local keypoints, computes rotationally robust 3D gradient-based descriptors, and aggregates them into image-level visual-word representations. These features are then used for low-dimensional visualization and logistic regression classification, while model weights are mapped back to the original volumes to generate attention maps that localize discriminative structures. We applied the pipeline to two cerebellar granule neuron datasets spanning both ideal and non-ideal imaging conditions. In a near-isotropic lattice light-sheet dataset of chromatin organization, the method separated control and NIPBL loss-of-function nuclei and supported accurate classification, with strongest performance in the facultative heterochromatin and H3.3 channels. Attention mapping and downstream connected-component and Haralick analyses revealed that loss-of-function nuclei contained more fragmented high-attention regions and smoother, more homogeneous chromatin-associated textures than controls. We then evaluated the same framework on an anisotropic confocal timelapse dataset of receptor clustering in dense neuronal cultures, where single-cell segmentation was impractical. Despite these challenges, the representation captured the expected ligand-driven clustering response and resolved subtler differences associated with a polarity protein overexpression. Together, these results establish a simple, interpretable, and broadly applicable framework for extracting biologically meaningful structure from volumetric microscopy datasets while preserving native 3D context.
Occupational burnout is common in the mental health care workforce, with negative consequences for professionals and patients. This study aimed to evaluate the efficacy of a digital health intervention to alleviate burnout in psychological therapists. This randomized controlled trial recruited 135 therapists working across 17 psychological services in England. The intervention involved six online group webinars based on principles of job crafting. Half of the participants accessed the intervention immediately (Group 1), and half were assigned to a waitlist control group (Group 2). After 6 weeks, Group 2 started the intervention. Participants completed measures of burnout (primary outcome), well-being, and job satisfaction at four time points (baseline, 6, 12, 36 weeks). Outcomes were compared between groups using mixed-effects models controlling for baseline severity and clustering by service. Differences between groups were statistically significant after 6 weeks, favoring job crafting versus waitlist control in burnout (d = 0.43, p < .001), well-being (d = -0.39, p = .023), and job satisfaction (d = -0.28, p = .006) measures. However, the magnitude of improvements relative to baseline levels declined over a 36-week period. A brief job crafting intervention led to short-term improvements in occupational health indicators. (PsycInfo Database Record (c) 2026 APA, all rights reserved).
The aim of this paper is to accurately assess the epistemological status of Albertus Magnus' (ca 1200-80) alchemy. In modern bibliography there is a "black or white" approach to the question of whether Albertus regarded alchemy as an art or scientia, and thus the adoption of an absolute thesis on the matter tends to create a series of interpretational problems. In contrast, I argue that Albertus' approach to alchemy does not exclude either art or scientia and, depending on the context, the Dominican master sometimes considers alchemy as an art-when it is connected with manual labor-and sometimes as a scientia-when it is connected to natural-philosophical and metaphysical aspects. At the end of my paper, I offer a way of properly assessing the Albertian alchemy, and determining how and when to accurately connect it with the notions of art and scientia.
In the oil industry, undesirable events in oil wells can cause economic losses, environmental accidents, and human casualties. In 2019, recognizing the importance and the lack of public datasets related to undesirable events in oil wells, Petrobras developed and publicly released the first version of the 3W Dataset, which is essentially a set of Multivariate Time Series labeled by experts. Since then, the 3W Dataset has been developed collaboratively and has become a foundational reference for numerous works in the field. This data article describes the current publicly available version of the 3W Dataset, which contains additional instances, more variables, and a new label. Furthermore, a new data structure has been developed to make data access more robust and efficient. The detailed description we provide encourages and supports the 3W community and new 3W users to improve previous published results and to develop new robust methodologies, digital products and services capable of detecting undesirable events in oil wells with enough anticipation to enable corrective or mitigating actions.
As infectious diseases spread, governments and specialised agencies typically release a main message conveying the recommended preventive action, accompanied by side messages justifying the action. In this study, we explored how side messages influence intentions to engage in infection prevention behaviours, attitudes toward people who do versus do not support infection prevention measures, and evaluations of the poster. We conducted three repeated cross-sectional online surveys in Japan and the United States in September, November, and December 2024. Participants were presented with posters promoting hand washing and mask wearing that included one of five side messages: self/other protection, public health, social norms, self/other protection plus norms, or public health plus norms. Meta-analyses of the three surveys showed side-message effects on intentions varied by country and baseline compliance: in Japan, for individuals who initially had high compliance tendencies, side messages combining social norms with additional justifications (self/other protection or public health) promoted behavioural intent, whereas in the United States, side messages that justified public health outcomes were most effective in promoting behavioural intent. We found no evidence that side-message content altered intergroup bias, defined as the difference in attitudes towards people who share one's own views versus those holding opposing views regarding infection prevention measures. However, the posters were perceived as coercive and stressful, particularly among participants with low baseline compliance, indicating that campaign designers may consider crafting messages that encourage adherence to infection prevention measures while taking care not to increase criticism of non-compliant individuals, who are often in the minority.
S-adenosyl-l-methionine (SAM)-dependent methyltransferases (MTs) are generally classified as C-, O-, N-, S-, or halide MTs depending on their methyl acceptor. C-MTs catalyze selective methylation reactions of carbon nucleophiles and play a crucial role in the regulation and diversification of natural products. The control of chemoselectivity by these enzymes is poorly understood, especially with respect to the resonance of a nucleophilic neighboring group that activates the carbon methylation site. We investigated two aromatic C-MTs for the underlying mechanisms governing their chemo- and/or regioselectivity. The unprecedented in vitro dimethylation activity of SfmM2 and NapB5 was demonstrated using the native substrate l-tyrosine and substrates with a 2,4-dihydroxyacetophenone pattern, respectively. Substrate symmetry and the in situ SAM supply with removal of the competitive inhibitor S-adenosyl-l-homocysteine are favorable for dimethylation activity. Through NapB5 catalysis, we obtained C-(di-)methylated acetylphloroglucinol and flavonoid derivatives. We discovered that NapB5 catalyzes both C- and O-methylation of sterically demanding flavonoids. Here, chemoselectivity was modulated by the geometry of substrate binding through substrate selection or site-directed mutagenesis. Precise positioning of the acceptor nucleophile toward SAM is required to achieve regio- and chemoselectivity despite competing C- and O-nucleophilic sites. Thus, chemoselectivity is context-dependent, which opens new horizons for the diversification of natural products.
Vulnerable road users in informal urban environments confront a distinct set of hazards that standard computer vision datasets are ill-equipped to represent: artisanal speed bumps constructed without regulatory compliance, deteriorated road markings, and the mototaxi-a three-wheeled motorized vehicle that constitutes the primary informal transport mode in intermediate Andean cities yet is absent from all major international repositories. This paper presents QHAWAY-from Quechua qhaway, a transitive verb meaning "to look; to observe"-an Advanced Driver Assistance System (ADAS) predicated on instance segmentation, monocular distance estimation via the pinhole camera model, and Time-to-Collision (TTC) computation, developed for the road environment of Ayacucho, Peru (2761 m a.s.l.), a city recognised by UNESCO as a Creative City of Crafts and Folk Art since 2019. A hybrid dataset comprising 25,602 images with 127,525 annotated instances across 12 classes was assembled by combining an original local collection of 4598 images (10,701 instances) captured through four complementary acquisition methods across the five urban districts of the Huamanga province with three established international datasets (BDD100K, BSTLD, RLMD; 21,004 images, 116,824 instances). A three-phase progressive training strategy with monotonically increasing resolution (640, 800, and 1024 pixels) was evaluated as an ablation study. A multi-architecture comparison spanning YOLOv8L-seg and the YOLO26 family (nano, small, large) identified YOLO26L-seg as the best-performing model, attaining mAP50 Box of 0.829 and mAP50 Mask of 0.788 at epoch 179. The integration of ByteTrack multi-object tracking with the pinhole equation D=(Hreal×f)/hpx delineates operational risk zones aligned with the NHTSA forward collision warning standard (danger: <3 m; caution: 3-7 m; TTC threshold ≤ 2.4 s). The system sustains processing rates of 19.2-25.4 FPS on an NVIDIA RTX 5080 GPU. A systematic field survey established that 96% of the audited speed bumps fail to comply with MTC Directive No. 01-2011-MTC/14, constituting the first quantitative record of informal road infrastructure non-compliance in the Andean region. Validation was conducted under naturalistic driving conditions without staged scenarios. Grad-CAM explainability analysis, encompassing three complementary visualisation algorithms (Grad-CAM, Grad-CAM++, and EigenCAM), confirmed that model attention concentrates consistently on safety-critical objects.
Today, consumers' growing demand for higher-quality beer and their desire for unique sensory experiences are propelling the craft beer movement into a new trend within the beer industry's production and consumption landscape. Unlike traditional beers with their single-ingredient combinations, craft beers innovatively incorporate special functional adjuncts. These adjuncts produce unique aromas and nutritional value during fermentation, primarily manifested in flavor compounds such as alcohols, esters, and organic acids. However, current craft beer production faces challenges like extended fermentation cycles and flavor instability, limiting the development of its distinctive characteristics. This paper is the first to focus on analyzing the relationship between raw material and adjuncts characteristics and flavor formation in craft beer, aiming to provide innovative insights for the transformation and upgrading of the traditional beer industry.
Karst ecosystems, 15% of Earth's land, are critical for soil erosion research due to their unique geology and hydrology. This study investigated the patterns and trends of soil erosion in global karst regions from 2000 to 2020 using the Integrated Valuation of Ecosystem Services and Trade-offs (InVEST) model combined with robust trend methods (Theil-Sen and Mann-Kendall). Correlation analysis and scenario analysis were employed to quantify the driving factors and evaluate the contributions of various factors. A global decline in erosion was found, with intense erosion in southwest Asia, southern Europe, and northwestern North America. Asia had the highest soil erosion rate, followed by North America, Europe, and other regions. Among them, the soil erosion rate in North America showed an upward trend. Russia, China, and Europe were key in erosion reduction (29.98%, 28.01%, 18.68%). Rainfall strongly correlated with erosion; vegetation's link varied by region. Temperature negatively correlated with erosion in some areas. Scenario analysis quantified contributions of human activities (47.06%) and climate (26.68%). These findings highlight the joint role of natural and human factors in soil erosion management, important for crafting effective conservation strategies in karst areas globally.