Search Results (122)

Background/Objectives: Coagulation factor XIa (FXIa) has emerged as a promising therapeutic target for the development of safer anticoagulant therapies with reduced bleeding risk. This study aimed to develop an interpretable machine learning-driven quantitative structure–activity relationship (QSAR) framework for predicting the inhibitory activity of FXIa inhibitors and supporting virtual screening applications. Methods: A total of 3026 curated compounds retrieved from the ChEMBL database were used for regression modeling, whereas 2119 compounds were retained for classification modeling after excluding intermediate-activity molecules. Molecular descriptors were generated using RDKit, Mordred, and Morgan fingerprint representations. Following preprocessing and feature selection, multiple machine learning algorithms were systematically benchmarked. Model robustness and reliability were further evaluated using 5-fold cross-validation, scaffold-aware validation, applicability domain analysis, and Y-randomization testing. Results: Nonlinear ensemble learning approaches consistently outperformed conventional linear algorithms. The optimized HistGradientBoostingRegressor achieved the best regression performance, with an independent test-set R² value of 0.711 and an RMSE value of 0.759, whereas the optimized classification model achieved accuracies approaching 95%. SHAP analysis identified lipophilicity-related descriptors, aromatic scaffold organization, electrostatic surface properties, and molecular topology as major contributors to FXIa inhibitory activity prediction. In addition, a proof-of-concept virtual screening workflow successfully identified several candidate compounds exhibiting high predicted pKi values and elevated active-class probabilities. Conclusions: The proposed framework provides a robust, interpretable, and reproducible machine learning-driven QSAR strategy for FXIa inhibitor discovery and may facilitate future virtual screening campaigns and medicinal chemistry optimization studies targeting FXIa-associated anticoagulant drug discovery. Full article

Collision avoidance in maritime unmanned aerial vehicle (UAV) operations must satisfy two competing objectives: ensuring reliable safety separation and minimizing unnecessary maneuver commands that increase operator burden and communication overhead. While deep reinforcement learning (DRL) has shown promise in handling high-dimensional encounter states, standard DRL approaches often prioritize safety at the cost of operational suitability, leading to frequent, oscillatory, or unnecessary avoidance commands that erode remote operator trust and consume limited communication bandwidth. To address this challenge, this paper proposes Resource-Aware Intrinsic Surprise Exploration (RAISE), a unified framework that balances collision avoidance performance with command economy. We conceptualize the issuance of avoidance maneuvers as a consumable “virtual resource”, compelling the agent to optimize its intervention budget. RAISE integrates this mechanism into the Soft Actor–Critic (SAC) architecture, augmented by a surprise-based intrinsic reward derived from the ensemble forward dynamics prediction error. This allows the agent to efficiently explore complex encounter scenarios driven by curiosity, while a resource-aware coefficient adaptively suppresses redundant actions when the communication or operational budget is constrained. Furthermore, an adaptive exponential moving average (EMA) scaling mechanism is introduced to stabilize the interplay between intrinsic and extrinsic rewards. Extensive simulations under diverse resource constraints and encounter geometries demonstrate that RAISE outperforms state-of-the-art baselines. It significantly reduces maneuver reversal rates and strengthens command stability without compromising safety margins. Specifically, under resource-constrained settings, RAISE suppresses excessive and unstable advisory behavior by reducing strengthening and reversal commands while maintaining effective collision avoidance; under resource-rich settings, it flexibly enhances safety buffers, demonstrating superior adaptability and operational realism for autonomous maritime UAV systems. Robustness evaluation confirms that RAISE maintains stable performance under sensor noise and wind disturbances. Full article

The rapid global expansion of photovoltaic capacity, now exceeding 1 TW, has transformed solar power forecasting from an engineering problem into a financially critical investment decision. Yet virtually all published forecasting studies optimise statistical accuracy metrics without translating improvements into monetised operational value. This paper introduces a unified cost–benefit framework that maps forecast errors across three operationally distinct time horizons onto imbalance costs, arbitrage revenues, and AI deployment costs. The economic conclusions are grounded in Romanian Balancing Market conditions (mean up-regulation price λ⁺ ≈ 85 €/MWh, mean down-regulation price λ⁻ ≈ 42 €/MWh; 15 min settlement interval), a five-year dataset (2018–2022) from a 10 MW utility-scale PV installation in Romania, and an annual AI system cost of 36,000 €/MW decomposed into data infrastructure, cloud GPU compute, and model-monitoring personnel. A Temporal Fusion Transformer ensemble, benchmarked against CNN-LSTM, Informer, and smart-persistence baselines, achieves a 0.38 Skill Score at the day-ahead horizon and a 0.28 Value Score, translating to a net economic benefit of €142,000 per installed MW per annum after full AI system cost deduction. While the framework is designed to be reusable across markets, all reported economic values are specific to the stated Romanian market parameters and should be recalibrated for other regulatory jurisdictions. Full article

Computer-aided drug discovery increasingly depends on virtual-screening workflows that remain reliable under severe class imbalance, chemical redundancy and early-recognition constraints. In this study, we developed a leakage-aware prioritization workflow for two cancer-relevant targets, pyruvate kinase M2 (PKM2) and mitogen-activated protein kinase 1 (MAPK1/ERK2), using the LIT-PCBA benchmark. The workflow combines canonical-SMILES curation, duplicate and label-conflict auditing, scaffold-aware validation, a non-learning nearest-active Tanimoto baseline, imbalance-aware machine-learning models, repeated-seed robustness analysis, isotonic probability calibration, ensemble-disagreement estimation, absorption, distribution, metabolism, excretion and toxicity (ADMET)-aware triage, molecular docking, and residue-level contact analysis. Benchmark enrichment is interpreted alongside calibration, ADMET filtering, docking and residue-contact evidence, rather than as a standalone discovery claim. PKM2 emerged as the clearer machine-learning case, with scaffold-aware tree models improving early recognition beyond the nearest-active similarity baseline and yielding top-ranked candidates supported by calibrated activity scores, ADMET profiles, docking scores, and residue-contact fingerprints. MAPK1 provided a biologically relevant contrast target, where ligand-neighborhood similarity remained competitive and downstream structural triage became more decisive than ligand-based ranking alone. These results support a conservative drug-discovery workflow in which leakage-aware benchmarking, calibration, uncertainty, and molecular-level triage remain visible throughout candidate prioritization. Full article

High-throughput single-cell RNA sequencing (scRNA-seq) has substantially advanced plant transcriptional landscapes. However, decoding cell-type-specific transcriptional regulation in non-model crops like tomato (Solanum lycopersicum) remains challenging. An integrated computational pipeline was applied using high-dimensional weighted gene co-expression (hdWGCNA) and ensemble machine learning to analyze tomato leaf single-cell transcriptomes. Unsupervised clustering identified 19 cell subpopulations mapped to five major cell-types: mesophyll cells (50.6%), guard cells (31.0%), trichomes (8.3%), vascular cells (7.5%), and lamina epidermis (2.6%). hdWGCNA revealed eight cell-type-specific modules, linking mesophyll cells to photosynthesis and guard cells to redox homeostasis. Machine learning classifiers prioritized candidate transcription factors (TFs), with XGBoost achieving the highest accuracy (0.85) to define cell identity. A consensus of 33 core TFs was identified, from which four candidate TFs (SlWRKY-78, SlWRKY-75, SlERF-57, and SlGLK-49) were selected for in silico knockout (KO) analysis. The simulations predicted that these knockouts might dysregulate core functional pathways, such as serine-type endopeptidase inhibitor activity and protein binding. Furthermore, CellOracle simulations suggested that the virtual deletion of the guard-cell-associated SlWRKY-78 and SlWRKY-75 could induce a directional trajectory shift from the terminally differentiated guard cells back to the less differentiated mesophyll territory. These findings provide a promising computational framework for deciphering cell-type-specific regulatory programs in horticultural crops. Full article

In roll-to-roll (R2R) web processing systems, traction rollers impose precise velocity profiles on the moving web. Ideally, the web follows this trajectory without deviation, but slip can occur during rapid acceleration or deceleration, leading to tension loss and degraded product quality. Although slip can be detected directly using high-resolution encoders that track the actual web speed, such sensors are expensive and require machine downtime for installation, making them impractical for large-scale industrial deployment. To overcome this limitation, we developed a virtual slip sensor that estimates slip using existing machine signals only. A temporary encoder was used to collect ground-truth data, enabling the training of predictive models that eliminate the need for a permanent physical sensor. The proposed system employs an ensemble modeling approach: a CatBoost model captures low-slip behavior where data is abundant, while a linear model extrapolates to high-slip, out-of-distribution conditions. Targeted feature engineering ensures generalization across varying ramp times and web speeds. Despite being trained primarily on data containing limited slip, the models successfully generalized to scenarios with severe slip, demonstrating robust predictive performance. The ensemble reduces the regular CatBoost model’s MSE at 60 m/min by approximately 54% in the speed-based evaluation and by approximately 68% in the quantile-based evaluation while maintaining comparable performance in the low-speed regimes. The resulting virtual sensor enables continuous real-time slip monitoring, providing operators with timely insights to prevent quality degradation and operate at higher acceleration profiles to increase throughput, even on machines that have not previously experienced extreme slip. Full article

Computational chemistry has played a central role in early-stage drug discovery by accelerating target selection, hit identification, and lead optimization. This review summarizes recent developments in molecular docking, pharmacophore modeling, molecular dynamics (MD), and virtual screening (VS), with a focus on their application in practical drug discovery workflows. Advances in docking protocols, including consensus scoring, physics-based rescoring, and ensemble approaches, addressed the challenges of receptor flexibility. Both ligand-based and structure-based pharmacophore models facilitated scaffold hopping and guided library prioritization. MD simulations were used to assess binding pose stability, identify cryptic binding pockets, and characterize solvent interactions. These simulations also supported free-energy calculations using endpoint and alchemical methods. Large-scale VS campaigns employed curated compound libraries, often composed of make-on-demand molecules, and relied on high-performance computing or cloud infrastructure to screen up to 10⁹ compounds. Hits were validated using orthogonal biophysical assays and filtered by absorption, distribution, metabolism, excretion, and toxicity (ADMET) predictions. Integrated pipelines combining pharmacophore modeling, docking, MD, and free-energy calculations improved enrichment rates and reduced the number of compounds requiring synthesis. Several case studies demonstrated the identification of nanomolar-affinity leads from ultra-large screening campaigns. The review also addressed ongoing challenges, such as inconsistent scoring of binding affinity, protonation, and tautomeric errors, dataset bias, and reproducibility issues. Strategies to mitigate these limitations included standardized library preparation, adherence to FAIR (Findable, Accessible, Interoperable, and Reusable) data principles, and the use of prospective benchmarking protocols. The review discussed emerging trends, including the use of quantum chemistry for electronic structure refinement, ensemble docking guided by cryo-electron microscopy (cryo-EM) data, and the integration of computational tools with automated synthesis and high-throughput screening in closed-loop discovery systems. These approaches have the potential to accelerate the design–make–test cycle, increase hit novelty, and improve decision-making in early drug development programs. Full article

Background/Objectives: Cyclooxygenase-2 (COX-2) is a well-established target in the development of anti-inflammatory drugs due to its central role in mediating inflammation. The identification of novel COX-2 inhibitors remains a key focus in pharmaceutical research. This study aimed to develop a robust and interpretable machine learning framework to predict COX-2 inhibitory activity and support virtual screening efforts. Methods: A curated dataset of 2052 compounds was obtained from the ChEMBL database. Molecular structures were encoded using Morgan fingerprints derived from SMILES representations. Several machine learning algorithms were trained and evaluated, including ensemble-based methods. Model performance was assessed using internal validation and external test sets. Robustness was further evaluated through Y-randomization tests. Model interpretability was investigated using SHAP (SHapley Additive exPlanations) analysis to identify key structural features contributing to activity. Results: Among the evaluated models, ensemble methods demonstrated superior predictive performance, with the Random Forest algorithm providing the most consistent and reliable results across validation and external datasets. Y-randomization confirmed that the model predictions were not due to chance correlations. SHAP analysis revealed that the most influential features corresponded to chemically meaningful substructures aligned with known COX-2 pharmacophore characteristics. The final optimized model was successfully deployed as a publicly accessible web application for real-time prediction using SMILES input. Conclusions: This study demonstrates the effectiveness of explainable machine learning approaches in predicting COX-2 inhibitory activity. The developed framework provides a reliable and interpretable tool for accelerating COX-2 inhibitor discovery and facilitating virtual screening in drug development. Full article

Despite its continued growth, cloud computing remains susceptible to significant security challenges, as shared virtualised environments pose threats at multiple levels. These vulnerabilities are caused by a lack of security coverage in the responsibility model between the provider and the tenant. In this work, we propose the multi-layered architecture VIRTUOSO (VIRTual Unified Operation Security Optimiser) to cover these security gaps through advanced automation and ML. VIRTUOSO has four layers. The Input Layer extracts key risk components from collected telemetry data. The Deep Automation Security Layer provides automated actions and continuous monitoring of security defences. Its counterpart, the Intelligent Security Layer, predicts threats using anomaly detection. The last layer, the Output Layer, returns an aggregated risk summary. The datasets we used were chosen for their relevance: the UNSW-NB15 dataset, a subset of the web-attack classification from CSE-CIC-IDS2018, and a sample of anonymised log events from AWS CloudTrail. Our ensemble classifiers achieve a best accuracy of 95.08% ± 0.13% on UNSW-NB15 (RF), with statistically significant differences among models confirmed by the Friedman test (p < 0.004) and Nemenyi post hoc analysis, and 99.25% ± 0.52% on web-attack (CatBoost), where ensemble differences are not statistically significant (p = 0.093), consistent with the high separability of this dataset. The training-test gap and DNN curves show no overfitting, whereas our adversarial tests show a maximum accuracy loss of 8.1% at ε = 0.02. With these promising results, we can assert that, pending verification in an actual cloud environment and potential integration with FL, our ensemble classifier model appears to be a good real-world prototype. Full article

Wind energy is a cornerstone of Saudi Arabia’s renewable energy transition under Vision 2030, yet national-scale wind resource assessment remains constrained by sparse and unevenly distributed ground observations. This study evaluates the spatiotemporal variability of near-surface wind speed and direction across Saudi Arabia using Global Ensemble Forecast System Reforecast (GEFSv12 Reforecast) wind fields integrated with a GIS 10.8-based processing workflow. Wind vectors (U and V) were extracted from NetCDF files, converted to wind speed and meteorological wind direction, and analyzed at 183 grid-cell “virtual stations” covering the Kingdom for a five-year period (2018–2022) at four synoptic time steps (6-hourly). The resulting database comprises approximately 1,336,632 records. A practical verification using five airport stations matched to nearest virtual stations shows strong agreement between GEFS-derived and observed wind speeds (RMSE = 1.823; R² = 0.879), supporting the dataset’s suitability for regional screening. Results reveal pronounced spatial heterogeneity and diurnal structure: northern, northeastern, central, and eastern Saudi Arabia consistently exhibit moderate-to-high winds (often >5.5 m/s) with persistent northwesterly–westerly flow, while western and southwestern coastal zones show stronger diurnal variability associated with thermal and sea-breeze influences. Peak, spatially coherent winds occur during the late-day synoptic period, forming a broad high-wind corridor across central and eastern regions. Given the ~1° (~110 km) resolution, findings are intended to be used for macro-scale wind-resource screening and the prioritization of high-wind zones for follow-up assessment. Full article

The increasing risk of air pollution in closed areas like passenger vehicles requires smart and real-time air quality reading solutions. Gases such as carbon monoxide (CO)—which is colorless and odorless and is produced by exhaust systems—air conditioners, and combustion sources are very dangerous to health because they can cause respiratory distress and poisoning at high levels. Traditional in-vehicle CO monitoring systems use a single-point sensor and a fixed threshold, which are insufficient in a dynamic cabin environment subject to factors such as vehicle size, ventilation rate, number of occupants, and incoming traffic. To address these drawbacks, this paper proposes a new E-Nose system with Virtual Sensor-Augmented Embedded Intelligence to estimate the CO concentration in vehicle cabins in real time. The system combines data from cheap gas sensors and improves it using virtual sensor machine learning models trained to predict or enhance sensor responses in real time. Embedded intelligence, deployed locally on edge hardware, supports low-latency processing, dynamic calibration, and noise filtering to respond to fluctuating environmental conditions adaptively. This architecture enables more accurate, robust, and context-aware estimation of CO levels compared to traditional threshold-based methods. Experimental validation across varied vehicular scenarios demonstrates superior precision and responsiveness, providing timely warnings even under complex dispersion patterns. Classifier Gradient Boosting, which builds an ensemble of weak learners sequentially, matched the Random Forest with 99.94% training and 98.59% model accuracy, confirming its strong predictive capability. The system is designed to be cost-effective, scalable, and easily integrable into modern automotive platforms. This study also contributes to the field of smart ecological recording and demonstrates the effectiveness of the virtual sensor-enhanced embedded system as an effective way to improve passenger safety by providing pre-emptive on-board air quality monitoring. Full article

Industrial pouring processes operate under highly dynamic conditions where small deviations can lead to defects, scrap, and production losses. Although modern foundries are equipped with multiple sensors and visual inspection systems, most monitoring approaches remain fragmented, unimodal, and difficult to interpret. Furthermore, annotated anomalous samples in industrial settings are scarce, hindering the development of traditional methods. As a result, many critical pouring anomalies are detected too late or lack sufficient contextual information for effective decision making. In this work, we propose a multimodal framework for industrial scene characterisation that combines visual information and process signals through an explainable Mixture-of-Experts (MoE)-style expert-fusion strategy. First, we deploy an ensemble of specialised modules that collaborate to identify regions of interest, assess pouring quality, and contextualise events within the production process, thereby generating an interpretable description of pouring events. Second, we introduce a novel anomaly detection method for multimodal video data, combining a self-supervised transformer with an outlier-aware clustering algorithm. Our approach effectively identifies rare anomalies without requiring extensive manual labelling. The resulting information is structured into a digital twin-ready representation, supporting synchronisation between the physical system and its virtual counterpart. This solution provides a scalable, deployable pathway to transform heterogeneous industrial data into actionable knowledge, supporting advanced monitoring, anomaly detection, and quality control in real foundry environments. Full article

Background: Breast cancer (BC) is one of the most diagnosed malignancies and a leading cause of cancer-related mortality among women worldwide, thereby posing a substantial threat to women’s health worldwide. However, clinically robust diagnostic biomarkers with high sensitivity and specificity, as well as well-validated molecular targets for targeted therapy, remain limited. Methods: BC transcriptomic data from seven GEO datasets and the TCGA-BRCA cohort (n = 1231) were integrated for analysis. After batch-effect correction, candidate genes were screened through DEA, WGCNA, and PPI networks analysis. An ensemble machine learning (ML) framework incorporating 127 algorithmic combinations was constructed, and SHAP analysis was applied to identify hub genes. Further analyses included functional enrichment, immune infiltration, miRNA regulatory network analysis, and SMR analysis. The expression patterns were validated using single-cell transcriptome data. Drug repositioning analysis and AI-assisted virtual screening were performed to prioritize compounds with favorable drug-like properties. The predicted binding modes of candidate compounds with CHEK1 were assessed by molecular docking. Results: Thirty core genes were obtained through differential expression, WGCNA, and PPI screening. Integrated ML (127 algorithms) determined the optimal model (AUC = 0.919), and SHAP identified nine feature genes, among which CHEK1 and KIF23 showed preliminary diagnostic potential across four external cohorts (AUC: 0.625–0.938). Functional enrichment indicated that both are enriched in the cell cycle and p53 pathways, closely associated with BRCA1/ATR; immune infiltration revealed significant correlations with macrophages and CD8⁺ T cells, with hsa-miR-15a-5p and hsa-miR-607 being common upstream regulatory miRNAs. SMR analysis supported a causal relationship between CHEK1 expression and BC genetic susceptibility (p_SMR < 0.05, p_HEIDI > 0.05); single-cell analysis confirms its heterogeneous expression. AI-assisted virtual screening identified 25 A-grade computational candidate compounds from 171 candidates. Molecular docking suggested that Olaparib and LY294002 can form favorable interactions with the CHEK1 active pocket. Conclusions: The study identified CHEK1 as a key diagnostic gene for BC through 127 ML algorithms and SMR causal inference. By combining AI-assisted virtual screening and molecular docking, computational candidate compounds targeting CHEK1 were prioritized. These findings represent hypothesis-generating in silico predictions and require experimental validation before any therapeutic conclusions can be drawn. Full article

Urban flood peak mitigation by green infrastructure (GI) is fundamentally a timing problem. Because GI storage is finite, interception occurs only within a brief active window; whether it reduces the outlet peak depends on GI placement in the network, routing lags, and rainfall timing. Here, we develop a timescale-based framework that links outlet peak reduction to the alignment among within-storm temporal structure, network response, and GI filling dynamics, providing a compact way to interpret when different network positions become most effective under a fixed GI design. Starting from a general convolution representation of runoff generation, interception, and routing, we show that peak reduction efficiency and location ranking can be organized by two nondimensional ratios—comparing storm duration and network response time to a characteristic GI filling time—plus simple descriptors of within-storm temporal structure. Under uniform rainfall, these ratios yield an interpretable regime diagram with analytical transition curves between downstream-, mid-network-, and upstream-optimal placement for a generic dispersive routing representation. Relaxing the uniform-rainfall assumption shows that within-storm variability can substantially reorganize these regimes because storm timing controls both how long GI storage remains available before it fills and which routed contributions overlap to form the outlet peak. Highly concentrated storms and storms with early internal peaks are especially likely to reorder the ranking of candidate locations relative to the uniform-rainfall baseline. Using 2351 observed hourly storm events evaluated across virtual catchments spanning fast to slow network responses, we quantify how often realistic event structure alters the optimal location and the regret associated with adopting a uniform design storm. The results motivate robustness-oriented placement strategies based on ensembles of plausible storm temporal structures, organized within the proposed timescale diagram rather than reliance on a single design hyetograph. Full article

Protein kinase inhibition can be achieved through various mechanisms, including blocking phosphorylation activity or disrupting regulatory interactions. While small molecule inhibitors have shown promise, their selectivity remains challenging due to the structural similarities among kinase catalytic sites. To design selective kinase inhibitors based on peptide terminal tail interactions with the activation segment, focusing on five kinases with different conformational states: GSK3, PAK4, TTN (OUT conformation) and PKB, FLT3 (IN conformation). Three-dimensional structures from RCSB PDB were optimized using MODELLER version 9.0. Peptide sequences were designed with PeptiDerive (Rosetta) and RosettaDesign version 3.5, followed by pharmacophore modeling based on key interaction residues. Virtual screening was then conducted with PyRx 0.8 and molecular docking with AutoDock Vina 1.1.2. Molecular dynamics simulations were performed using Desmond v6.6 (Schrödinger Suite 2016, Multisim v3.8.5.19) (100 ns, NPT ensemble, 300 K). Analysis of the five kinases revealed distinct interaction profiles with designed peptidomimetic compounds. Kinases displaying the IN conformation of the activation segment (PKB and FLT3) consistently showed superior stability and stronger interaction profiles compared to those in the OUT conformation. The designed compounds formed key hydrogen bonds and hydrophobic interactions with critical residues in the activation segment binding pocket. The most promising inhibitors demonstrated stability throughout the molecular dynamics simulations, with IN conformation kinases maintaining more consistent conformational profiles than their OUT conformation counterparts. Kinases with IN conformation of the activation segment demonstrated superior stability and interaction profiles compared to OUT conformations. These findings contribute to our understanding of selective kinase inhibition and provide a framework for developing novel inhibitors, particularly for PKB and FLT3. The implications of this study extend to rational drug design approaches that leverage natural regulatory mechanisms for therapeutic intervention, though further optimization is needed for GSK-3β, PAK4, and TTN to improve stability and binding affinity. Full article