Workflow For Prediction Research Articles

Deep learning-based overall survival prediction in patients with glioblastoma: An automatic end-to-end workflow using pre-resection basic structural multiparametric MRIs

PurposeAccurate and automated early survival prediction is critical for patients with glioblastoma (GBM) as their poor prognosis requires timely treatment decision-making. To address this need, we developed a deep learning (DL)-based end-to-end workflow for GBM overall survival (OS) prediction using pre-resection basic structural multiparametric magnetic resonance images (Bas-mpMRI) with a multi-institutional public dataset and evaluated it with an independent dataset of patients on a prospective institutional clinical trial. Materials and methodsThe proposed end-to-end workflow includes a skull-stripping model, a GBM sub-region segmentation model and an ensemble learning-based OS prediction model. The segmentation model utilizes skull-stripped Bas-mpMRIs to segment three GBM sub-regions. The segmented GBM is fed into the contrastive learning-based OS prediction model to classify the patients into different survival groups. Our datasets include both a multi-institutional public dataset from Medical Image Computing and Computer Assisted Intervention (MICCAI) Brain Tumor Segmentation (BraTS) challenge 2020 with 235 patients, and an institutional dataset from a 5-fraction SRS clinical trial with 19 GBM patients. Each data entry consists of pre-operative Bas-mpMRIs, survival days and patient ages. Basic clinical characteristics are also available for SRS clinical trial data. The multi-institutional public dataset was used for workflow establishing (90% of data) and initial validation (10% of data). The validated workflow was then evaluated on the institutional clinical trial data. ResultsOur proposed OS prediction workflow achieved an area under the curve (AUC) of 0.86 on the public dataset and 0.72 on the institutional clinical trial dataset to classify patients into 2 OS classes as long-survivors (>12 months) and short-survivors (<12 months), despite the large variation in Bas-mpMRI protocols. In addition, as part of the intermediate results, the proposed workflow can also provide detailed GBM sub-regions auto-segmentation with a whole tumor Dice score of 0.91. ConclusionOur study demonstrates the feasibility of employing this DL-based end-to-end workflow to predict the OS of patients with GBM using only the pre-resection Bas-mpMRIs. This DL-based workflow can be potentially applied to assist timely clinical decision-making.

Advanced Mechanical Analysis Workflow for Chiplets Integration Predictive Design

Chiplet-based designs offer a compelling value proposition compared to traditional monolithic System on Chip (SoC) architectures with advantages in IP reuse, performance, cost, and time to market. Mechanical modeling and analysis (M&amp;A) play a critical role in enabling the integration of chiplets, spanning the chip-package-board-system domains to support product development. The integration of chiplets from various suppliers and deverse technologies introduces differences in material properties, thermal coefficients, and mechanical behaviors that presents a significant challenge in ensuring a good mechanical reliability across the chiplet-based integration system. Consequently, chiplet-based designs encounter more challenges to address the elevated mechanical stress and reliability concerns. Traditional mechanical analysis and simulation workflows are primarily tailored for monolithic SoC-based designs and are not readily applicable to chiplet-based designs. The latter can no longer afford to overlook issues related to mechanical and thermal stress induced by multiple chiplets, especially with the use of advanced 2.5D and high-density fanout packages. In this paper, an advanced mechanical analysis workflow for chiplet-based designs is proposed. The representative examples that demonstrate the efficacy of this workflow are also presented. The advanced mechanical analysis workflow involves modeling of chiplets interactions at the early design stage spanning from the Si level to package level and encompassing scales from micrometers to millimeters that includes the components such as Through-Silicon Vias (TSVs), HIgh Density Fanout Cu RDL (Redistribution Layer) and microbumps. Furthermore, the predictive mechanical analysis workflow extends to analyses at the millimeter to meter scale, considering entire systems and incorporating elements like heat sinks and printed circuit boards (PCBs). It represents a multifaceted process spanning chip-to-package co-design, stress analysis, thermal analysis, failure analysis, and chiplet-based design optimization.

Transforming Physiology and Healthcare through Foundation Models.

Recent developments in artificial intelligence (AI) may significantly alter physiological research and healthcare delivery. Whereas AI applications in medicine have historically been trained for specific tasks, recent technological advances have produced models trained on more diverse datasets with much higher parameter counts. These new, "foundation" models raise the possibility that more flexible AI tools can be applied to a wider set of healthcare tasks than in the past. This review describes how these newer models differ from conventional task-specific AI, which relies heavily on focused datasets and narrow, specific applications. By examining the integration of AI into diagnostic tools, personalized treatment strategies, biomedical research, and healthcare administration, we highlight how these newer models are revolutionizing predictive healthcare analytics and operational workflows. In addition, we address ethical and practical considerations associated with the use of foundation models by highlighting emerging trends, calling for changes to existing guidelines, and emphasizing the importance of aligning AI with clinical goals to ensure its responsible and effective use.

Low-Code BPM meets IoT: A Framework for Real-Time Industrial Automation

Abstract:As we enter the age of Industry 4.0 with smart factories, these interconnected systems, industrial automation has come a long way. To enable real time automation of industrial processes, this paper proposes a novel framework for combining Low-Code Business Process Management (BPM) platforms with the Internet of Things (IoT) technologies. Low Code BPM brings about agility and ease of development as well with IoT offering real time data from sensors and devices closing the gap between generation of data and getting actionable insights. It addresses critical problems including response latency, process inefficiency and scalability limits. The study shows substantial improvements in response time by integrating IoT data streams with Low Code BPM workflows, from 12x faster, to better process efficiency from 70% to 92% and robust scalability to 100 devices with minimal latency. Middleware, edge computing, and intuitive interfaces were used to mitigate challenges such as device integration complexity, data overload and users adaptability. This research serves to support Industry 4.0 goals of predictive maintenance, workflow optimization, and dynamic decisions making. Keywords: Low-code BPM, IoT integration, real-time automation, industry 4.0, smart manufacturing, predictive maintenance, workflow optimization

Large scale Raman spectrum calculations in defective 2D materials using deep learning

We introduce a machine learning prediction workflow to study the impact of defects on the Raman response of 2D materials. By combining the use of machine-learned interatomic potentials, the Raman-active Γ-weighted density of states method and splitting configurations in independant patches, we are able to reach simulation sizes in the tens of thousands of atoms, with diagonalization now being the main bottleneck of the simulation. We apply the method to two systems, isotopic graphene and defective hexagonal boron nitride, and compare our predicted Raman response to experimental results, with good agreement. Our method opens up many possibilities for future studies of Raman response in solid-state physics.

BenchAMRking: a Galaxy-based platform for illustrating the major issues associated with current antimicrobial resistance (AMR) gene prediction workflows

BackgroundThe Joint Programming Initiative on Antimicrobial Resistance (JPIAMR) networks ‘Seq4AMR’ and ‘B2B2B AMR Dx’ were established to promote collaboration between microbial whole genome sequencing (WGS) and antimicrobial resistance (AMR) stakeholders. A key topic discussed was the frequent variability in results obtained between different microbial WGS-related AMR gene prediction workflows. Further, comparative benchmarking studies are difficult to perform due to differences in AMR gene prediction accuracy and a lack of agreement in the naming of AMR genes (semantic conformity) for the results obtained. To illustrate this problem, and as a capacity-building exercise to encourage stakeholder involvement, a comparative Galaxy-based BenchAMRking platform was developed and validated using datasets from bacterial species with PCR-verified AMR gene presence or absence information from abritAMR.ResultsThe Galaxy-based BenchAMRking platform (https://erasmusmc-bioinformatics.github.io/benchAMRking/) specifically focusses on the steps involved in identifying AMR genes from raw reads and sequence assemblies. The platform currently comprises four well-characterised and published workflows that have previously been used to identify AMR genes using WGS data from several different bacterial species. These four workflows, which include the ISO certified abritAMR workflow, make use of different computational tools (or tool versions), and interrogate different AMR gene sequence databases. By utilising their own data, users can investigate potential AMR gene-calling problems associated with their own in silico workflows/protocols, with a potential use case outlined in this publication.ConclusionsBenchAMRking is a Galaxy-based comparison platform where users can access, visualise, and explore some of the major discrepancies associated with AMR gene prediction from microbial WGS data.

AI-Powered Fintech innovations for credit scoring, debt recovery, and financial access in Microfinance and SMEs

The integration of artificial intelligence in fintech is revolutionizing financial services, particularly for microfinance institutions and small and medium-sized enterprises (SMEs). This paper explores the transformative impact of AI-powered innovations in credit scoring, debt recovery, and financial access. AI-driven credit scoring leverages alternative data and advanced machine learning techniques to enhance accuracy, inclusivity, and efficiency, addressing the limitations of traditional methods. In debt recovery, AI optimizes collection processes through predictive analytics, workflow automation, and conversational tools, improving operational efficiency while fostering ethical practices and customer trust. AI also plays a pivotal role in expanding financial access, enabling underserved populations to benefit from tailored digital platforms for lending, savings, and insurance. Despite its potential, AI adoption entails risks, including data privacy concerns, algorithmic bias, and the digital divide, which require careful management. The paper concludes with recommendations for policymakers, financial institutions, and tech developers to ensure AI's ethical and inclusive deployment, fostering economic resilience and equitable growth. Keywords: Artificial Intelligence (Ai), Fintech Innovations, Microfinance Institutions (Mfis), Small And Medium-Sized Enterprises (Smes), Credit Scoring, Financial Inclusion.

Mandibular Reconstruction with Osseous Free Flap and Immediate Prosthetic Rehabilitation (Jaw-in-a-Day): In-House Manufactured Innovative Modular Stackable Guide System.

Background: Head and neck reconstruction following ablative surgery results in alterations to maxillofacial anatomy and function. These postoperative changes complicate dental rehabilitation. Methods: An innovative modular, stackable guide system for immediate dental rehabilitation during mandibular reconstruction is presented. The virtual surgical planning was performed in Materialise Innovation Suite v26 and Blender 3.6 with the Blenderfordental add-on. The surgical guides and models were designed and manufactured at the point of care. Results: The duration of the surgery was 9 h and 35 min. Good implant stability (>35 Ncm) and a stable occlusion were achieved. After 9 months of follow-up, the occlusion remained stable, and a mouth opening of 25 mm was registered. The dental implants showed no signs of peri-implant bone loss. Superposition of the preoperative planning and postoperative position of the fibula parts resulted in an average difference of 0.70 mm (range: -1.9 mm; 5.4 mm). Conclusions: The in-house developed stackable guide system resulted in a predictive workflow and accurate results. The preoperative virtual surgical planning was time-consuming and required extensive CAD/CAM and surgical expertise. The addition of fully guided implant placement to this stackable guide system would be beneficial. More research with longer follow-ups is necessary to validate these results.

Monthly Prediction of Pine Stress Probability Caused by Pine Shoot Beetle Infestation Using Sentinel-2 Satellite Data

Due to the significant threat to forest health posed by beetle infestations on pine trees, timely and accurate predictions are crucial for effective forest management. This study developed a pine tree stress probability prediction workflow based on monthly cloud-free Sentinel-2 composite images to address this challenge. First, representative pine tree stress samples were selected by combining long-term forest disturbance data using the Continuous Change Detection and Classification (CCDC) algorithm with high-resolution remote sensing imagery. Monthly cloud-free Sentinel-2 images were then composited using the Multifactor Weighting (MFW) method. Finally, a Random Forest (RF) algorithm was employed to build the pine tree stress probability model and analyze the importance of spectral, topographic, and meteorological features. The model achieved prediction precisions of 0.876, 0.900, and 0.883, and overall accuracies of 89.5%, 91.6%, and 90.2% for January, February, and March 2023, respectively. The results indicate that spectral features, such as band reflectance and vegetation indices, ranked among the top five in importance (i.e., SWIR2, SWIR1, Red band, NDVI, and NBR). They more effectively reflected changes in canopy pigments and leaf moisture content under stress compared with topographic and meteorological features. Additionally, combining long-term stress disturbance data with high-resolution imagery to select training samples improved their spatial and temporal representativeness, enhancing the model’s predictive capability. This approach provides valuable insights for improving forest health monitoring and uncovers opportunities to predict future beetle outbreaks and take preventive measures.

SmartGraph API: Programmatic Knowledge Mining in Network-Pharmacology Setting.

The recent SmartGraph platform facilitates the execution of complex drug-discovery workflows with ease in the network-pharmacology paradigm. However, at the time of its publication we identified the need for the development of an Application Programming Interface (API) that could promote biomedical data integration and hypothesis generation in an automated manner. This need was magnified at the time of the COVID-19 pandemic. This study addresses the absence of such an API. Accordingly, most functionalities of the original platform were implemented within the SmartGraph API. We demonstrate that by using the API it is possible to transform the original semiautomated workflow behind the Neo4COVID19 database to a fully automated one. The availability of the SmartGraph API lends a significant improvement to the programmatic integration of network-pharmacology-oriented knowledge graphs and analytics, as well as predictive functionalities and workflows.

Accelerated Exploration of Empty Material Compositional Space: Mg-Fe-B Ternary Metal Borides.

Borides are a family of materials with valuable properties for various applications. Their diverse structures and compositions, yet disparity in the constituent chemical elements for the known compounds, give elemental substitutions for prototypes great potential for material discovery. To explore uncharted material compositional space, we develop a workflow that joins high-throughput crystal structure prediction and automated diffraction pattern matching to discover new compounds with significant prediction and synthesis hurdles. Utilizing the workflow, we explore the empty Mg-Fe-B ternary compositional space, previously uncharted largely due to the immiscibility of Mg and Fe, as a paradigm. A total of 275 ternary boride prototypes are classified, using which we predict 23 (158) stable and metastable ternary phases within 50 (200) meV/atom above the convex hull. We identify Gd2(FeB)7-type Mg2Fe7B7 and ZrCo3B2-type MgFe3B2 to match previously unsolved experimental powder X-ray diffraction (PXRD) patterns. The discovered Mg2Fe7B7 and related channeled structures feature mismatched Mg and (FeB) sublattice periods, for which we conduct structural analyses with respect to the PXRD. They are predicted to exhibit exceptionally fast superionic transport of Mg and outstanding electrochemical performance, which serve as post-Li-ion battery candidate electrode materials. This result opens a new avenue for borides' potential applications as electrode materials and fast ionic conductors. This work also portrays the map and landscape of ternary metal borides with similar chemical environments. With high efficiency, the prototype- and PXRD-assisted crystal structure prediction workflow opens a new avenue for exploring various material compositional spaces across the periodic table.

Predictable Full Digital Workflow Using Stackable Surgical Templates for Complete Dental Arch Rehabilitation with Implant-Supported Fixed Restorations-Case Series and Proof of Concept.

In recent years, advancements in digital dentistry have provided new opportunities for more predictable and efficient treatment options, particularly in patients with failing dentition. This study aimed to evaluate the effectiveness and accuracy of a fully digital workflow using stackable surgical templates for complete dental arch rehabilitation with implant-supported fixed restorations. Four patients, comprising two males and two females with a mean age of 66 years, were included in this case series. Each patient underwent meticulous digital planning, including CBCT and intraoral scanning, to create a virtual patient for preoperative assessment and virtual treatment planning. The assessment of the trueness of implant positioning was conducted in Geomagic Control X software (version 2017.0.3) by referencing anatomical landmarks from both the preoperative and one-year postoperative CBCT scans. A total of 25 dental implants were placed in the maxilla, followed by the installation of long-term provisional restorations. The results showed minimal deviation between the planned and actual implant positions, with mean 3D coronal, apical, and angular discrepancies of 0.87 mm, 2.04 mm, and 2.67°, respectively. All implants achieved successful osseointegration, and no failures were recorded, resulting in a 100% survival rate at the one-year follow-up. Patients reported high satisfaction with both the esthetic and functional outcomes based on their subjective feedback. The findings suggest that the use of a fully digital workflow with stackable surgical templates is a reliable and effective approach for immediate implant placement and prosthetic rehabilitation, enhancing treatment precision and patient comfort.

Molli: A General Purpose Python Toolkit for Combinatorial Small Molecule Library Generation, Manipulation, and Feature Extraction.

The construction, management, and analysis of large in silico molecular libraries is critical in many areas of modern chemistry. Herein, we introduce the MOLecular LIibrary toolkit, "molli", which is a Python 3 cheminformatics module that provides a streamlined interface for manipulating large in silico libraries. Three-dimensional, combinatorial molecule libraries can be expanded directly from two-dimensional chemical structure fragments stored in CDXML files with high stereochemical fidelity. Geometry optimization, property calculation, and conformer generation are executed by interfacing with widely used computational chemistry programs such as OpenBabel, RDKit, ORCA, NWChem, and xTB/CREST. Conformer-dependent grid-based feature calculators provide numerical representation and interface to robust three-dimensional visualization tools that provide comprehensive images to enhance human understanding of libraries with thousands of members. The package includes a command-line interface in addition to Python classes to streamline frequently used workflows. Parallel performance is benchmarked on various hardware platforms, and common workflows are demonstrated for different tasks ranging from optimized grid-based descriptor calculation on catalyst libraries to an NMR chemical shift prediction workflow from CDXML files.

An end-to-end multi-task network for early prediction of the instrumental intensity and magnitude in the north–south seismic belt of China

The seismic activity in the north–south seismic belt of China is among the highest in the world. Predicting instrumental intensity and magnitude after an earthquake mitigates regional seismic disasters. The standard workflow for prediction involves building empirical formulas using characteristic parameters of the initial arrival seismic wave, but this method has limitations in accuracy. Recent data-driven models have shown promise in predicting instrument intensity and magnitude. Still, this is currently done mainly on a single-task basis and does not consider whether a multi-task model can utilize complementary information from different tasks to improve overall performance. This study proposes a data-driven multi-task model called SeismNet, which can simultaneously predict instrument intensity and magnitude. We tested the effectiveness of SeismNet using ground motion records of the north–south seismic belt of China. The model can predict instrument intensity and magnitude more rapidly and accurately than the baseline and single-task models, with increasing accuracy as the input seismic wave duration increases. We also tested the method on three destructive earthquake events (Ms > 6.5) that occurred in China and found that at 3 s after the P-wave arrival, the prediction is almost consistent with the observation. Overall, this study offers a new method for improving earthquake prediction accuracy in the North-South seismic belt of China.

Development and validation of a machine learning model integrated with the clinical workflow for inpatient discharge date prediction.

Discharge date prediction plays a crucial role in healthcare management, enabling efficient resource allocation and patient care planning. Accurate estimation of the discharge date can optimize hospital operations and facilitate better patient outcomes. In this study, we employed a systematic approach to develop a discharge date prediction model. We collaborated closely with clinical experts to identify relevant data elements that contribute to the prediction accuracy. Feature engineering was used to extract predictive features from both structured and unstructured data sources. XGBoost, a powerful machine learning algorithm, was employed for the prediction task. Furthermore, the developed model was seamlessly integrated into a widely used Electronic Medical Record (EMR) system, ensuring practical usability. The model achieved a performance surpassing baseline estimates by up to 35.68% in the F1-score. Post-deployment, the model demonstrated operational value by aligning with MS GMLOS and contributing to an 18.96% reduction in excess hospital days. Our findings highlight the effectiveness and potential value of the developed discharge date prediction model in clinical practice. By improving the accuracy of discharge date estimations, the model has the potential to enhance healthcare resource management and patient care planning. Additional research endeavors should prioritize the evaluation of the model's long-term applicability across diverse scenarios and the comprehensive analysis of its influence on patient outcomes.

Identification of Magnaporthe oryzae candidate secretory effector proteins through standardizing the filtering process of the canonical parameters

AbstractThe virulence of Magnaporthe oryzae largely hinges on its secretory effectors. Therefore, identification and thorough understanding of the effector functionality is crucial for unravelling the pathogenicity of the pathogen. In the present study, we employed a modified computational pipeline with deep machine learning techniques with an integration of Magnaporthe effector reference datasets (MOED) that predicted 434 M. oryzae candidate secretory effector proteins (MoCSEPs) from the genomic data. The reliability of the modified CSEP prediction workflow through utilization of precise parametric filtering is considered valid as it predicted 100 functional effectors (97.08%) out of 103 previously identified effector proteins within the Magnaporthe genus. Insights into secretion patterns and subcellular localization elucidated the role of these proteins in host cell recognition. Furthermore, structural classification of MoCSEPs, based on conserved motifs, combined with an exploration of their biological functions, revealed their significance in host adaptability and localization. Experimental validation done through examining expression of the MoCSEPs revealed varied secretion patterns in the resistant (40 expressed) and susceptible (92 expressed) rice cultivars at different time intervals after pathogen inoculation owing to different degrees of resistance by the host cultivars. The present work thus provides the strategic model of canonical parametric evaluation within the MOED and deepens the understanding on the role of secretory proteins of M. oryzae in establishing successful parasitic infection in rice. The predicted MoCSEPs could be used as biomarkers for disease diagnosis and tracking evolutionary shifts in M. oryzae.

MADE-for-ASD: A multi-atlas deep ensemble network for diagnosing Autism Spectrum Disorder

In response to the global need for efficient early diagnosis of Autism Spectrum Disorder (ASD), this paper bridges the gap between traditional, time-consuming diagnostic methods and potential automated solutions. We propose a multi-atlas deep ensemble network, MADE-for-ASD, that integrates multiple atlases of the brain’s functional magnetic resonance imaging (fMRI) data through a weighted deep ensemble network. Our approach integrates demographic information into the prediction workflow, which enhances ASD diagnosis performance and offers a more holistic perspective on patient profiling. We experiment with the well-known publicly available ABIDE (Autism Brain Imaging Data Exchange) I dataset, consisting of resting state fMRI data from 17 different laboratories around the globe. Our proposed system achieves 75.20% accuracy on the entire dataset and 96.40% on a specific subset — both surpassing reported ASD diagnosis accuracy in ABIDE I fMRI studies. Specifically, our model improves by 4.4 percentage points over prior works on the same amount of data. The model exhibits a sensitivity of 82.90% and a specificity of 69.70% on the entire dataset, and 91.00% and 99.50%, respectively, on the specific subset. We leverage the F-score to pinpoint the top 10 ROI in ASD diagnosis, such as precuneus and anterior cingulate/ventromedial. The proposed system can potentially pave the way for more cost-effective, efficient and scalable strategies in ASD diagnosis. Codes and evaluations are publicly available at https://github.com/hasan-rakibul/MADE-for-ASD.

Multi‐view street view image fusion for city‐scale assessment of wind damage to building clusters

AbstractGlobal warming amplifies the risk of wind‐induced building damage in coastal cities worldwide. Existing numerical methods for predicting building damage under winds have been limited to virtual environments, given the prohibitive costs associated with establishing city‐scale window inventories. Hence, this study introduces a cost‐effective workflow for wind damage prediction of real built environments, where the window inventory can be established with the multi‐view street view image (SVI) fusion and artificial intelligence large model. The feasibility of the method is demonstrated based on two real‐world urban areas. Notably, the proposed multi‐view method surpasses both the single‐view and aerial image‐based methods in terms of window recognition accuracy. The increasing availability of SVIs opens up opportunities for applying the proposed method not only in disaster prevention but also in environmental and energy topics, thereby enhancing the resilience of cities and communities from multiple perspectives.

Geometric deep learning for molecular property predictions with chemical accuracy across chemical space

Chemical engineers heavily rely on precise knowledge of physicochemical properties to model chemical processes. Despite the growing popularity of deep learning, it is only rarely applied for property prediction due to data scarcity and limited accuracy for compounds in industrially-relevant areas of the chemical space. Herein, we present a geometric deep learning framework for predicting gas- and liquid-phase properties based on novel quantum chemical datasets comprising 124,000 molecules. Our findings reveal that the necessity for quantum-chemical information in deep learning models varies significantly depending on the modeled physicochemical property. Specifically, our top-performing geometric model meets the most stringent criteria for “chemically accurate” thermochemistry predictions. We also show that by carefully selecting the appropriate model featurization and evaluating prediction uncertainties, the reliability of the predictions can be strongly enhanced. These insights represent a crucial step towards establishing deep learning as the standard property prediction workflow in both industry and academia.Scientific contributionWe propose a flexible property prediction tool that can handle two-dimensional and three-dimensional molecular information. A thermochemistry prediction methodology that achieves high-level quantum chemistry accuracy for a broad application range is presented. Trained deep learning models and large novel molecular databases of real-world molecules are provided to offer a directly usable and fast property prediction solution to practitioners.

Direct operational data-driven workflow for dynamic voltage prediction of commercial alkaline water electrolyzers based on artificial neural network (ANN)

The thermal inertia of alkaline water electrolysis (AWE) system under dynamic operation reduces efficiency and poses security risks. To mitigate this, predictive model control (MPC) emerges as a potential solution, aiming to precisely regulate temperature through temperature trend prediction, which relies on accurate voltage predictions. However, the current MPC implementation for AWE system utilizes empirical model for voltage prediction, requiring extensive experimental data for calibration, making it time-consuming and costly for large-scale water electrolysis projects. Data-driven models are a promising alternative for voltage prediction, but research is limited by the lack of long-term commercial AWE operational data and the ambiguity in developing effective data-driven models. In this study, a 10-kW level AWE test system underwent 300 h of simulated solar-driven dynamic operation to assess the resilience of AWE system to dynamic conditions. Subsequently, a robust data-driven workflow using artificial neural networks (ANNs) was developed for precise dynamic voltage prediction based on obtained operational data. The data-driven workflow incorporates automated processes for voltage prediction, including data importing, data cleaning, ANN architecture optimization, model training, and ultimately voltage prediction, avoiding the need for manual parameters fitting. The model’s performance was evaluated and compared with empirical model across four dynamic operation types (cold-start, ramp, stepwise, and random response), achieving an impressive accuracy over 98 %. Furthermore, the ANN model also exhibits application potential in U-I curves prediction (accuracy > 98 %), influential features assessment, operational optimization and MPCs. This ANN-based workflow serves as a versatile framework for precise voltage prediction under dynamic operations and provides an analytical platform for operational data analysis, system optimization, and advanced control system development.