Skip to main content
NCBI home page
ACS AuthorChoice logoLink to ACS AuthorChoice
. 2026 May 29;130(27):5282–5292. doi: 10.1021/acs.jpca.6c01154

IPSBrain: A Unified Intelligent Data Analysis Platform for Multimodal Experimental Characterization at Advanced Photon Source

Yihe Pang †,‡, Lin Li †,§, Xiangwen Deng †,#, Zuoxin Xi †,∥, Jing Li †,∥, Mengdan Han †,∥, Yaping Sun †,⊥, Haodong Yao †,#, Liyuan Wu †, Qin Yuan †, Lina Zhao †,#,*
PMCID: PMC13359377  PMID: 42214078

Abstract

Advanced photon sources deliver X-ray with unprecedented brilliance and coherence, facilitating sophisticated multimodal characterization of diverse materials while simultaneously generating data volumes that far exceed the capacity of traditional, expert-driven analytical workflows. This growing gap between data acquisition and interpretation has become a critical bottleneck for timely scientific discovery, which creates a critical need for high-throughput and automated data analysis tools. Recent advances in artificial intelligence (AI) offer a transformative potential to address this challenge, enabling the direct mapping of complex experimental observations into scientific insights. However, existing AI-driven tools are remain largely fragmented and task-specific, limiting their adoption of modern multimodal experiments at the facility scale. Here we present IPSBrain, a unified AI-powered platform that serves as Intelligent Photon Source Brain for automated analysis of multimodal synchrotron experimental data. The platform integrated a series of novel, self-developed machine-learning-based models that enable automated data analysis from multiple experimental techniques, including diffraction, scattering, X-ray absorption, and tomograph. The platform is accessible through an intuitive web interface, which significantly lowers the technical barrier to advanced scientific data interpretation. By enabling end-to-end intelligent workflows from data acquisition to quantitative physical insight, IPSBrain supports high-throughput, reproducible, and scalable analysis aligned with the operational demands of modern synchrotron radiation sources. IPSBrain not only demonstrates a practical and extensible framework for AI-driven synchrotron data analysis but also establishes a transferable paradigm for intelligent data analysis of other large-scale research infrastructures, such as neutron sources and free-electron lasers.

graphic file with name jp6c01154_0009.jpg

graphic file with name jp6c01154_0007.jpg

Introduction

Synchrotron radiation is the electromagnetic radiation emitted when charged particles, such as electrons, are accelerated in a magnetic field. Characterized by ultrahigh brightness, broad spectral coverage, high coherence, pulsed time structure, and tunable polarization, synchrotron radiation has become a powerful, nondestructive probe for resolving material structure and dynamics across scales from macroscopic features to atomic dimensions. Photon sources are large-scale facilities designed to generate and utilize such radiation efficiently, which have evolved through successive generations, from early facilities derived from particle colliders to second-generation sources built specifically for synchrotron experiments, followed by third-generation sources based on insertion devices such as undulators and wigglers.

With the emergence of the fourth-generation photon source, the brightness and coherence of synchrotron radiation have increased by orders of magnitude over previous generations. , These capabilities have pushed structural characterization to unprecedented spatial and temporal resolution, fundamentally reshaping research paradigms across materials science, structural biology, chemistry, physics, geoscience, and cultural heritage studies. − The advent of fourth-generation synchrotron sources has been accompanied by rapid advances in detector technologies, driving a paradigm shift in experimental data volume from gigabyte (GB) to terabyte (TB) scales per experiment. At the same time, the diversification of experimental techniques (e.g., X-ray diffraction and scattering, absorption spectroscopy, and imaging) has led to inherently multimodal data sets. , These massive, high-dimensional, and complex scientific data contain essential information for uncovering novel physical phenomena, biological processes, and advanced material properties. However, accurate and efficient extraction of key physical insights from such vast, multimodal data has emerged as a critical bottleneck limiting the efficiency and depth of scientific discovery enabled by large-scale infrastructures.

Traditional methods for analyzing experimental data rely heavily on domain expertise and are typically performed through manual procedures or the sequential use of multiple specialized software tools. Such approaches are difficult to standardize and automate, resulting in data processing speeds far behind data acquisition rates, creating a critical efficiency bottleneck. To address these challenges, the integration of artificial intelligence (AI), particularly machine learning and deep learning, has emerged as a promising trend in synchrotron data analysis. − Recent efforts have begun to leverage artificial intelligence technology to analyze data acquired from synchrotron radiation facilities, aiming to overcome the gap between data collection and interpretation. For example, dedicated machine learning tools such as BraggNN, PtychoNN, and Ptychography have been developed to accelerate Bragg-peak localization and coherent diffraction imaging (CDI) reconstruction, enabling near real-time data analysis at beamlines of the Advanced Photon Source (APS). In parallel, the National Synchrotron Light Source II (NSLS-II) developed an X-ray crystallography-assisted agent (XCA), which employed a trained probabilistic model to identify crystal phases in experimental measurements, enabling rapid analysis of high-throughput X-ray diffraction data. Despite these advances, these task-specific tools often operate in isolation and lack interoperability, thereby motivating the development of integrated, platform-level solutions. The Extensible Workflow System (EWOKs) at ESRF offers a web-based platform for seamlessly integrating diverse existing data analysis software and flexibly customizing data processing workflows from raw data reduction to publication-ready results. The Virtual Infrastructure for Scientific Analysis (VISA) platform provides a cloud-based environment that allows users to launch virtual machines for accessing experimental data and executing workflows for photon and neutron science. Besides, the MLExchange focuses on AI-driven data analysis, enabling scientists with limited machine learning (ML) expertise to train, test, and deploy ML models seamlessly within scientific facilities. Nevertheless, current software and platforms remain task-specific tools or manually defined workflows, which limit efficient multimodal analysis and broader adoption, particularly under the data-intensive demands of fourth-generation light sources, motivating the need for an AI-native unified platform capable of automated data interpretation across diverse synchrotron experiments.

Here, we present a platform, IPSBrain, which serves as the Intelligent Photon Source Brain for automated analysis of multimodal synchrotron experimental data. The platform integrates dedicated self-developed machine learning algorithms designed for the high-dimensional, large-scale scientific data and combines automated processing pipeline with interactive visualization. As shown in Figure , by unifying all analysis services and applications for diffraction, scattering, absorption spectroscopy, and computed tomography images within an intuitive web-based interface, the platform substantially lowers the barrier to advanced data interpretation. The platform is publicly accessible at https://www.ipsbrain.com/, allowing free use without restrictions. Its development and deployment enable rapid, high-throughput data analysis and extraction of key physical information from complex experiments, thereby accelerating the transformation of synchrotron radiation facilities, especially China’s High Energy Photon Source (HEPS), toward automated and intelligent operation.

1.

1

Open in a new tab

The IPSBrain platform for intelligent analysis of multimodal synchrotron data. Schematic overview of the IPSBrain platform, illustrating its integrated framework for the automated analysis and processing of multimodal synchrotron radiation data. The platform supports three main synchrotron radiation experiments: X-ray diffraction and scattering, X-ray absorption spectroscopy, and tomographic imaging, enabling high-throughput, standardized, and quantitative AI-assisted data analysis across complementary experimental modalities.

Methods

Platform Implementation

IPSBrain adopts a cloud-native, microservice-based architecture designed to meet requirements of robust, efficient, and extensible online analysis for multimodal and massive synchrotron radiation experimental data.

As shown in Figure , the platform is organized into a hierarchical architecture with decoupled front-end and back-end components: the interface layer is used for interactive application access; the gateway layer provides secure and high-throughput access by combining load balancing and dynamic routing transfer to sustain heavy concurrent traffic, while enforcing TLS encryption and unified authentication to ensure data integrity and user privacy; the core service layer adopts a microservice architecture that integrates system services and AI-driven analytical algorithms, enabling independent deployment, elastic scaling, and reliable service governance. Underlying these components is an infrastructure layer that integrates persistent storage, distributed indexing, high-performance caching, and asynchronous messaging, all supported by a dedicated high-performance computing cluster. More details about the platform’s core components are introduced in the following.

2.

2

Open in a new tab

Architecture diagram of the IPSBrain platform.

Microservice Framework

To achieve system resilience, development agility, and operational efficiency, IPSBrain adopts a microservice architecture that modularizes functional components, allowing them to operate collaboratively while remaining independently deployable and scalable, thereby facilitating platform maintenance and iterative updates. As shown in Figure , “Service Deployment and Governance”, the IPSBrain organizes its functionality into two major categories of services: system services and AI algorithm services. The system services provide the core foundation for system operation, encompassing distributed task scheduling, message dissemination, data encryption, system monitoring, logging, and file management. The AI algorithm services include shared utilities such as task allocation, sandboxed execution environments, secure data transfer, and file preprocessing, while each analytical algorithm is packaged as an independent, language-agnostic microservice, performing data processing and analysis.

Real Time Processing

To support high-throughput experimental workflows, the IPSBrain platform is designed with a real-time data processing capability that integrates tightly with the online data acquisition system of the high-energy photon source (HEPS). As illustrated in Figure , the system enables efficient request scheduling, lightweight model inference, and scalable software and hardware infrastructures support for continuous experimental data streams. At the gateway layer, the load balancer and routing transfer modules are responsible for managing incoming analysis requests from the data acquisition system. The load balancer dynamically distributes requests across available computing services, while the routing transfer module directs tasks to the appropriate analysis pipelines according to data modality and processing requirements. This gateway-level scheduling mechanism ensures a stable request flow and prevents congestion when large volumes of experimental data are generated in real time. At the AI algorithm service layer, IPSBrain adopts a lightweight deployment strategy for model inference. The deployed models are optimized for efficient CPU execution, achieving an average processing latency of 12.5 ms per data sample on a single CPU core, which is sufficient to match the throughput of typical online experimental data acquisition streams. This lightweight design enables rapid inference without requiring dedicated GPU resources, thereby facilitating scalable deployment across multiple computing nodes. In addition, the software and hardware infrastructure layers provide further support for real-time performance. Components such as distributed index, distributed cache, and messaging proxy enable efficient data access, intermediate result caching, and asynchronous message communication between system modules. These distributed services, together with the high-performance computing servers deployed within the platform environment, ensure reliable resource allocation and scalable processing capacity for large-scale experimental data sets. Collectively, these architectural components guarantee the stable, real-time data analysis capability of the IPSBrain platform.

Data Management

The IPSBrain platform includes data management services that cover the entire data analysis lifecycle, including a unified integration strategy for heterogeneous data, high-performance data storage and retrieval, and secure data access. To handle diverse metadata formats generated across beamlines, the platform adopts a unified data representation independent of the beamline-specific metadata strategy. Experimental data sets originating from different acquisition formats (e.g., NeXus, HDF5) are standardized prior to analysis. Specifically, diffraction, scattering, and spectroscopy data are organized as numerical arrays, whereas imaging experiments are represented as two-dimensional image data sets. This strategy decouples beamline-specific heterogeneous experimental metadata formats from downstream data analysis services while maintaining a unified analysis interface for the AI-driven processing.

To ensure high throughput and efficient data access, the IPSBrain platform employs a high-performance storage architecture combining persistent databases and distributed indexing mechanisms. MySQL serves as the relational database for persistent transactional data, while Elasticsearch provides distributed indexing to achieve efficient data querying and read and write operations. In addition, Redis acts as an in-memory cache for accelerating frequent data access and reducing system latency.

Besides, to further ensure data security and integrity, IPSBrain adopts a token-based authentication framework using digitally signed JSON Web Tokens (JWTs), which are validated at the API gateway to enforce access control and traffic management, effectively protecting data against unauthorized access, DDoS attacks, and common injection attacks (see Figure Gateway layer). All data exchanges, including uploads, downloads, and API communications, are secured using TLS 1.2, which combines asymmetric key negotiation with symmetric encryption to ensure confidentiality, integrity, and forward secrecy, while QUIC-based congestion control is used to improve transmission efficiency and reduce latency. Moreover, all data stored in the platform are protected using Transparent Data Encryption (TDE), which employs Advanced Encryption Standard with a 256-bit key (AES-256) and a hierarchical key management framework. The master key (MK), securely maintained in an external Key Management Service (KMS) or Hardware Security Module (HSM), decrypts the data encryption keys (DEKs) upon authorized access, which in turn decrypt the data pages for application-level use. This design enforces strict cryptographic isolation between storage-level encryption and application access while remaining fully transparent to upstream applications, effectively mitigating data leakage from lost or compromised storage media and ensuring both robust storage security and a high compatibility with analytical workflows.

Application Interface

Intelligent data analysis serves as the core application of the IPSBrain platform, primarily supporting three major synchrotron radiation experimental modalities: diffraction and scattering, absorption, and imaging. It enables high-throughput data acquisition, automated analysis and processing, and result visualization.

As shown in Figure , the platform currently incorporates six deep learning-based analytical algorithms, including crystal structure interpretation from X-ray diffraction patterns, high-quality data reconstruction of small-angle X-ray scattering, valence state identification from X-ray absorption, and automated segmentation of nanoscale CT images. To ensure broad accessibility, IPSBrain adopts a graphical, task-oriented web interface. Users can access diverse tools through intuitive navigation panels and parameter controls, enabling the execution of complete analytical workflows via simple click- or drag-and-drop operations with no-code programming. This significantly lowers the barrier to entry while enhancing the efficiency and reproducibility of the synchrotron radiation data analysis.

3.

3

Open in a new tab

User-friendly application service and operation interface. (a) Services selection interface, (b) data ingestion, parameter setting, and results visualization dashboard.

The interactive front end of IPSBrain is implemented using Vue.js, a lightweight progressive JavaScript framework that extends standard web technologies with a declarative template system and a highly optimized core library for rapid interface rendering. Vue.js employs a virtual document object model (DOM) and a reactive data-binding mechanism based on the model-view-view model (MVVM) architectural pattern, ensuring that changes in application state are efficiently propagated to the view layer while minimizing direct DOM operations. This design provides a responsive and computationally efficient user interface that is well-suited for complex, data-intensive analytical workflows.

Data Analysis Algorithms

The IPSBrain platform integrates six novel, self-developed intelligent analysis methods, including MSCNN and TransXRD for structural analysis and attribute inference of diffraction spectrum; U-NetAPRC and QuantSAXS for high-fidelity processing and quantitative analysis of scattering data; CuXAS for analysis of local-structure characterization of absorption spectra; and CSUNet for segmentation and quantitative analysis of nanotomography images.

Diffraction and Scattering Spectrum Analysis

Synchrotron diffraction and scattering techniques employ the high brightness, highly coherent X-rays produced by synchrotron sources to probe the microscopic structure of matter. , By measuring the intensity distribution of the resulting coherent or incoherent scattering signals in reciprocal space, these experimental techniques enable the reconstruction and interpretation of structural information at the microscopic scale. − Owing to the exceptional properties of synchrotron radiation, diffraction and scattering have become indispensable techniques for resolving crystal structures, lattice defects, nanoscale architectures, and even the three-dimensional arrangements of biological macromolecules. − However, modern beamline experiments routinely generate multimillion data points per measurement, which far exceeds the practical limits of manual expert-driven analysis. − Traditional data analysis pipelines are built on physics-driven and segmented procedures, including data reduction (e.g., Fit2D or DAWN), peak identification, model fitting (e.g., GSAS-II, SASfit), structural modeling (e.g., VESTA), and Rietveld refinement. While these approaches are methodologically complete, they are inherently labor-intensive, heavily reliant on specialist intuition. These limitations increasingly hinder their applicability to high-throughput and high-dimensional data sets, underscoring the need for intelligent and integrated frameworks for automated analysis of diffraction and scattering spectrum data.

The IPSBrain platform establishes an end-to-end automated pipeline that integrates four specialized machine-learning algorithms for diffraction and scattering data interpretation, enabling high-throughput automated extraction of microstructural characteristics and macroscopic material properties from raw experimental data. For X-ray diffraction (XRD), we present a multiscale deep convolutional neural network, namely MSCNN that identifies the crystal system and space group symmetry from XRD patterns, achieving a Top-3 accuracy of 96.7%. Building on this structural recognition, a transfer-learning-based model, TransXRD , predicts material properties (bandgap energy) with a mean squared error of 0.29 eV. For X-ray scattering, we present U-NetAPRC to reconstruct high-fidelity small-angle X-ray scattering (SAXS) patterns that mitigate radiation damage artifacts in protein measurements, thereby providing high signal-to-noise ratio scattering patterns for subsequent analysis. Moreover, we further developed a quantitative SAXS analysis module QuantSAXS that enables automated, single-click extraction of key solution structure parameters, including the maximum particle dimension (Dmax) and radius of gyration (Rg). Implementation details of these algorithms are provided in Supporting Information.

Absorption Spectrum Analysis

X-ray absorption spectroscopy (XAS) at synchrotron light sources provides unique atomic-scale insight into the local structure, oxidation state, and electronic environment of absorbing elements, making it a core experimental technique across diverse fields such as catalysis, functional materials, and energy storage systems. − However, traditional XAS analysis remains heavily reliant on forward simulation and spectral fitting, which require computationally intensive simulations using packages such as FEFF or MXAN. These workflows involve numerous user-defined parameters, leading to the process not only being labor-intensive and subjective but also fundamentally limited in throughput. With the advent of advanced light sources, , experimental data rates far outpace the capacity of manual analysis, creating a critical bottleneck between measurement and physical insight.

The IPSBrain platform offers an intelligent “inverse-structure” inference solution that predicts local structural descriptors directly from X-ray absorption spectra. The core model named CuXAS integrates a multilayer perceptron (MLP) with a random forest (RF) to enable one-click estimation of key structural parameters, including coordination number (CN), bond length (BL), and radial distribution function (RDF) for the Cu (copper) elemental system. This method establishes a streamlined and fully automated pipeline from spectral input to quantitative structural resolution, achieving a correlation coefficient (R2) of 0.95 for coordination number and bond length identification, with a mean absolute error (MAE) of 0.134 for radial distribution function (RDF) prediction. More details of the model training and evaluation are provided in Supporting Information.

Tomography Image Analysis

Synchrotron-based X-ray tomography leverages the high brilliance and coherence of synchrotron radiation to achieve fidelity imaging, with its high-resolution variant, nanotomography (Nano-CT), extending three-dimensional visualization to the nanoscale. Nano-CT enables nondestructive, three-dimensional imaging at nanometer spatial resolution, allowing visualization of intracellular ultrastructure without staining or sectioning. However, quantitative analysis of intracellular nanoparticles from Nano-CT images remains highly challenging, as the raw tomographic volumes are often compromised by artifacts and noise that distort the signal of nanoparticles and hinder accurate quantification of their spatial distribution. Traditional analysis approaches rely on handcrafted filtering and artifact-suppression procedures that are computationally intensive and lack scalability, limiting the application in high-throughput data streams of modern synchrotron radiation facilities. ,

To this end, the IPSBrain platform integrates a deep learning-assisted analytical neural network, CSUNet , for Nano-CT data that performs automated, precise segmentation of cellular compartments and nanoparticles. The CSUNet achieves pixel-level identification of nuclei, cytoplasm, and nanoparticles, with a mean-intersection-overunion (mIoU) score of 99.93%. Building upon these segmentation outputs, the framework executes an automated quantitative pipeline to extract critical spatial distribution metrics, including the volume fraction of nanoparticles inside and outside cells, their distribution density within the cytoplasm and nucleus, and the minimum distances from particles to cellular or nuclear boundaries. This integrated approach substantially reduces manual intervention, accelerates the processing of Nano-CT images and delivers superior segmentation accuracy under low-contrast conditions compared to traditional filtering-based methods. For details of the framework, please refer to our previous publication.

Results and Discussion

Here, we examine two representative use cases: high-throughput structure determination from diffraction data and single-cell-resolution analysis of synchrotron-based tomographic images, to illustrate the practical relevance of IPSBrain in modern materials and nanoscience.

Inorganic Nonmetallic Crystal Structure Detection

Accurate identification of material structures is fundamental to understanding their physicochemical properties and functional performance. X-ray diffraction (XRD) stands as a primary technique for crystal structure characterization, enabling determination of phase composition, crystallographic symmetry, lattice parameters, and residual strain through the analysis of characteristic diffraction patterns. Here, we present a representative use case of crystal symmetry identification from XRD patterns using the IPSBrain platform, demonstrating a critical and well-established application in high-throughput structural analysis.

The procedure for crystal symmetry analysis is as follows: First, upload diffraction spectrum file(s) containing diffraction angle (2θ)–intensity pairs (Figure b-1). The platform supports high-throughput analysis, allowing batch processing of up to 100 data sets simultaneously per submission. Upon initiating the analysis by triggering the prediction function, the platform automatically processes the data using its embedded analytical algorithm. After computation, the results are displayed in the output panel, reporting the most probable crystal system and space group for each spectrum, along with Top-3 and Top-5 candidate predictions ranked by confidence (Figure b-1). In this representative application, two diffraction patterns were selected for structural analysis: a transition metal sulfate containing hydrazine and a new ternary phosphide. The platform identified the probable crystal systems and space groups as orthorhombic (Pccn) and hexagonal (P6m2), respectively. These results are fully consistent with the structures obtained from Rietveld refinement (Figure a), demonstrating the accessibility and the reliability of the platform for high-throughput crystal structure analysis.

4.

4

Open in a new tab

Example of automated crystallographic symmetry analysis of XRD patterns using the IPSBrain platform. (a) Input XRD patterns and the corresponding structural model derived from expert interpretation. (b) Workflow and output of the IPSBrain platform: simple file upload and a point-and-click operation, then the platform automatically delivers accurate crystallographic symmetry results.

To highlight the efficiency advantage of the AI-driven platform, we compared IPSBrain with a conventional diffraction data analysis workflow implemented using GSAS-II software. As shown in Table , IPSBrain enables direct prediction of the crystal system and space group from a single diffraction pattern with millisecond-level latency and full automation, whereas traditional workflows require multiple sequential processing steps and substantially longer analysis times.

1. Performance Comparison of IPSBrain with Traditional Software in Crystal Structure Symmetry Identification.

  time/data automation workflow
IPSBrain 12.5 ms fully automatic 1D diffraction pattern → MSCNN module → crystal systems and space groups
GSAS-II 15 min semiautomatic 1D diffraction pattern → peak detection → peak fitting → indexing → symmetry analysis → crystal systems and space groups
Open in a new tab

Nanoscale Computed Tomography Image Segmentation

Owing to the spatial nanoscale resolution, synchrotron-based X-ray nanocomputed tomography (NanoCT) enables nondestructive 3D reconstruction of ultrasmall metal nanoparticles within cells. In this use case, we demonstrate the use of the IPSBrain platform for sensitive identification and quantitative analysis of synchrotron radiation tomographic imaging data at single-cell resolution.

The analysis was performed on 100 central nano-CT slices of a human breast cancer cell (MCF-7). As shown in Figure , a compressed data set containing the image sequence along with corresponding nucleus and cytoplasm masks was uploaded to the platform. Upon execution of the inference operation, the input images are segmented automatically and the results are visualized in the display panel. Subsequently, quantitative analysis of intracellular HfO2 nanoparticles is performed using user-specified imaging scale and analysis range parameters, with the computed metrics presented as line plots in the quantitative analysis panel. Accurate characterization of three-dimensional (3D) distributions of drug molecules at the single-cell level is essential for elucidating cellular uptake pathways and drug action mechanisms. This use case demonstrates the applicability and significance of the platform for automated analysis of synchrotron tomographic data at single-cell resolution, providing a valuable tool for nanoparticle drug design and advancing mechanistic understanding of cellular drug interactions.

5.

5

Open in a new tab

Example of synchrotron radiation tomographic data analysis using the IPSBrain platform. The analytical procedure is as follows: First, the data file in TIF format is uploaded, upon which the analytical model is invoked. Second, the segmentation results are displayed in the “SampleDisplay” panel. Third, quantitative analysis is performed, yielding the uptake volume of nanoparticles along with their proportional distribution within the cytoplasm and the nucleus. Finally, a quantitative measurement of the penetration depth is provided.

Multimodal Synchrotron Data Joint Analysis Reveals Multiscale Structural Evolution During Phase Transition

To demonstrate the platform’s multimodal data analysis capabilities, we collected the XRD pattern and XAS spectrum for a copper sample undergoing a structural phase transformation, and then analyzed these two sets of data using the corresponding modules of the IPSBrain platform. The combined analysis results are summarized in Figure . As shown in Figure a, crystallographic symmetry analysis of the XRD patterns indicates that the Cu sample adopts a cubic structure with space group Fm3̅m in one state, while it transforms to a tetragonal structure with space group I4/mmm in the other state. These results reveal a clear change in the long-range crystallographic symmetry associated with the structural transformation. Complementary information on the local atomic environment was obtained from XAS analysis (Figure b). In the cubic phase, the fitted coordination number is 8.3 with an average first-shell bond length of 2.69 Å. After the structural transition to the tetragonal phase, the coordination number slightly decreases to 8.03, accompanied by a reduction of the average first-shell bond length to 2.63 Å. Taken together, the combined XRD and XAS results provide a coherent description of the structural evolution of the Cu sample across different length scales. The transition from cubic to tetragonal symmetry reflects a change in long-range crystallographic order, while the concurrent decrease in coordination number and bond length indicates a subtle rearrangement of the local atomic environment. These results demonstrate that multimodal analysis enables a more comprehensive characterization of structural transformations by simultaneously resolving global symmetry changes and local coordination variations.

6.

6

Open in a new tab

Example of integrated XRD–XAS analysis using the IPSBrain platform. (a) Crystallographic symmetry analysis of XRD patterns from two elemental Cu samples using the crystal symmetry identification module. (b) Structural analysis of XAS spectra from the same two elemental Cu samples using the XAS structural analysis module.

Conclusion and Future Developments

Fourth-generation synchrotron radiation facilities generate massive, high-dimensional, and multimodal experimental data, posing substantial challenges to automatic and high-throughput analysis. In this study, we present IPSBrain, an intelligent analysis platform for synchrotron radiation data that integrates novel in-house developed machine learning models for large-scale multimodal scientific data. The IPSBrain delivers modular web-based applications that perform end-to-end data analysis pipelines and interactive visualization, enabling efficient AI-accelerated analysis of diffraction, scattering, X-ray absorption spectroscopy, and tomographic imaging data. Currently, the platform provides six core applications, i.e., crystal structure and properties determination from the X-ray diffraction (XRD), high-fidelity data reconstruction and quantitative analysis of small-angle scattering (SAXS) in biological solutions, extraction of coordination number, bond length, and radial distribution function (RDF) from the X-ray absorption spectroscopy (XAS), and segmentation of cell compartments and nanoparticles in X-ray tomography images. The IPSBrain is available as an open-access platform via https://www.ipsbrain.com/, providing unrestricted, free access to its analytical tools from any location at any time. The unified, user-friendly interface and extensible architecture of IPSBrain will significantly lower the technical barrier to advanced data interpretation, enabling rapid extraction of physical insights and accelerating the transition toward automated and intelligent operation at modern synchrotron radiation facilities.

Looking ahead, the IPSBrain platform will evolve along three complementary directions. First, we will expand the analytical applications to cover a broader range of experimental modalities and scientific targets, with refinement of machine learning models for enhanced accuracy and a shift toward explainable AI methodologies. Second, we will deepen the integration of the data processing pipeline, aiming to establish an end-to-end pipeline from the detector-captured raw data to the interpretation of quantitative physical parameters. Third, we will optimize the user interaction by more intuitive interfaces and responsive visualization, ensuring efficient, interactive access for researchers across disciplines. Furthermore, the integrated and AI-driven paradigm of IPSBrain is transferable to other large-scale scientific infrastructures, such as spallation neutron sources and free-electron laser facilities. We plan to establish a collaborative ecosystem to support the development of AI-based analysis algorithms for synchrotron experiments. We will make models and source code integrated within IPSBrain available to researchers interested in contributing to the community upon request, enabling them to fine-tune existing models or develop new models tailored to specific experimental applications. Through the microservice architecture, these new models can be integrated with built-in mechanisms for version control and lifecycle management, thereby enabling continuous platform evolution driven by community contributions. We hope that this approach will facilitate intelligent data analysis across facilities, promote methodological integration, and accelerate scientific discovery, thereby enhancing the scientific production of next-generation large-scale research infrastructures.

Supplementary Material

jp6c01154_si_001.pdf (73.8KB, pdf)

Acknowledgments

The authors gratefully acknowledge Prof. Yuhui Dong for valuable discussions and suggestions regarding the development of the intelligent data analysis platform, and Dr. Haifeng Zhao for helpful discussions on the methodology of the absorption spectroscopy presented in this work. This work was supported by the National Key Research and Development Program of China (grant nos. 2021YFA1200904), the National Natural Science Foundation of China (grant nos. 12375326), the Innovation Program for IHEP (grant nos. E35457U210), the Directional institutionalized Scientific Research Platform relies on the Beijing Synchrotron Radiation Facility of the Chinese Academy of Sciences (grant no. JZHKYPT-2021-02), and the Strategic Priority Research Program of Chinese Academy of Sciences (grant nos. XDA0480600, XDA0530301).

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpca.6c01154.

  • Implementation details of data analysis algorithms (PDF)

∇.

Y.P. and L.L. contributed equally to this work.

The authors declare no competing financial interest.

References

  1. Balerna, A. ; Mobilio, S. . Introduction to synchrotron radiation. In Synchrotron Radiation: Basics, Methods and Applications; Springer, 2014; pp 3–28. [Google Scholar]
  2. Freund A. K.. Third-generation synchrotron radiation X-ray optics. Structure. 1996;4(2):121–125. doi: 10.1016/S0969-2126(96)00015-9. [DOI] [PubMed] [Google Scholar]
  3. Shin S.. New era of synchrotron radiation: fourth-generation storage ring. AAPPS Bulletin. 2021;31(1):21. doi: 10.1007/s43673-021-00021-4. [DOI] [Google Scholar]
  4. Chapman H. N.. Fourth-generation light sources. IUCrJ. 2023;10:246–247. doi: 10.1107/S2052252523003585. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Schöder S., Müller K., Tranchant L., Rouquié A., Gueriau P., Thoury M., Bérard E., Okbinoglu T., Berenguer F., Iacconi C.. et al. Heritage research at the PUMA beamline. Appl. Phys. A:Mater. Sci. Process. 2024;130(11):848. doi: 10.1007/s00339-024-08026-0. [DOI] [Google Scholar]
  6. Newham E., Corfe I. J., Brewer P., Bright J. A., Fernandez V., Gostling N. J., Hoffmann S., Jäger K. R., Kague E., Lovric G.. et al. The origins of mammal growth patterns during the Jurassic mammalian radiation. Sci. Adv. 2024;10(32):eado4555. doi: 10.1126/sciadv.ado4555. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Gruhl T., Weinert T., Rodrigues M. J., Milne C. J., Ortolani G., Nass K., Nango E., Sen S., Johnson P. J., Cirelli C.. et al. Ultrafast structural changes direct the first molecular events of vision. Nature. 2023;615(7954):939–944. doi: 10.1038/s41586-023-05863-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Kapfer M., Jessen B. S., Eisele M. E., Fu M., Danielsen D. R., Darlington T. P., Moore S. L., Finney N. R., Marchese A., Hsieh V.. et al. Programming twist angle and strain profiles in 2D materials. Science. 2023;381(6658):677–681. doi: 10.1126/science.ade9995. [DOI] [PubMed] [Google Scholar]
  9. Lawrence Bright E., Giacobbe C., Wright J. P.. Beam heating from a fourth-generation synchrotron source. Synchrotron Radiat. 2021;28(5):1377–1385. doi: 10.1107/S160057752100669X. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Zhe S., Ming L.. Beamlines and applications of the fourth-generation synchrotron radiation light source. Physics. 2024;53(2):80–88. doi: 10.7693/wl20240202. [DOI] [Google Scholar]
  11. Fu T., Wang Y., Zhang K., Zhang J., Wang S., Huang W., Wang Y., Yao C., Zhou C., Yuan Q.. Deep-learning-based ring artifact correction for tomographic reconstruction. Synchrotron Radiation. 2023;30(3):620–626. doi: 10.1107/S1600577523000917. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Liu Z., Sharma H., Park J. S., Kenesei P., Miceli A., Almer J., Kettimuthu R., Foster I.. BraggNN: fast X-ray Bragg peak analysis using deep learning. IUCrJ. 2022;9(1):104–113. doi: 10.1107/S2052252521011258. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Cherukara M. J., Zhou T., Nashed Y., Enfedaque P., Hexemer A., Harder R. J., Holt M. V.. AI-enabled high-resolution scanning coherent diffraction imaging. Appl. Phys. Lett. 2020;117(4):044103. doi: 10.1063/5.0013065. [DOI] [Google Scholar]
  14. Babu A. V., Zhou T., Kandel S., Bicer T., Liu Z., Judge W., Ching D. J., Jiang Y., Veseli S., Henke S.. et al. Deep learning at the edge enables real-time streaming ptychographic imaging. Nat. Commun. 2023;14(1):7059. doi: 10.1038/s41467-023-41496-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Maffettone P. M., Banko L., Cui P., Lysogorskiy Y., Little M. A., Olds D., Ludwig A., Cooper A. I.. Crystallography companion agent for high-throughput materials discovery. Nat. Comput. Sci. 2021;1(4):290–297. doi: 10.1038/s43588-021-00059-2. [DOI] [PubMed] [Google Scholar]
  16. De Nolf W., Huder L., Svensson O., Payno H., Autran P.-O., Broche L., Koumoutsos G., Götz A.. EWOKS: An Extensible WOrKflow System. Synchrotron Radiat. News. 2024;37(6):9–15. doi: 10.1080/08940886.2024.2432305. [DOI] [Google Scholar]
  17. Zhao, Z. ; Chavez, T. ; Holman, E. A. ; Hao, G. ; Green, A. ; Krishnan, H. ; McReynolds, D. ; Pandolfi, R. J. ; Roberts, E. J. ; Zwart, P. H. ; Yanxon, H. ; Schwarz, N. ; Sankaranarayanan, S. ; Kalinin, S. V. ; Mehta, A. ; Campbell, S. I. ; Hexemer, A. . MLExchange: A web-based platform enabling exchangeable machine learning workflows for scientific studies, 2022 4th Annual Workshop on Extreme-scale Experiment-in-the-Loop Computing; 2022; pp 10–15. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Alshuqayran, N. ; Ali, N. ; Evans, R. . A systematic mapping study in microservice architecture, IEEE 9th international conference on service-oriented computing and applications (SOCA) 2016; 2016; pp 44–51. [Google Scholar]
  19. DuBois, P. MySQL; Addison-Wesley, 2013. [Google Scholar]
  20. Gormley, C. ; Tong, Z. . Elasticsearch: the definitive guide: a distributed real-time search and analytics engine; O’Reilly Media, Inc, 2015. [Google Scholar]
  21. Carlson, J. Redis in action.; Simon and Schuster, 2013. [Google Scholar]
  22. Dierks, T. ; Rescorla, E. . transport layer security (TLS) protocol version 1 2008; Vol. 2. [Google Scholar]
  23. Langley, A. ; Riddoch, A. ; Wilk, A. ; Vicente, A. ; Krasic, C. ; Zhang, D. ; Yang, F. ; Kouranov, F. ; Swett, I. ; Iyengar, J. . et al. The quic transport protocol: Design and internet-scale deployment, Proceedings of the conference of the ACM special interest group on data communication; 2017; pp 183–196. [Google Scholar]
  24. Madyatmadja E. D., Hakim A. N., Sembiring D. J. M.. Performance testing on Transparent Data Encryption for SQL Server’s reliability and efficiency. J. Big Data. 2021;8(1):134. doi: 10.1186/s40537-021-00520-z. [DOI] [Google Scholar]
  25. Biryukov, A. ; Khovratovich, D. ; Nikolić, I. . Distinguisher and related-key attack on the full AES-256, Annual International Cryptology Conference; Vol. 5677, 2009; pp 231–249 10.1007/978-3-642-03356-8_14. [DOI] [Google Scholar]
  26. Wong, C. K. ; Lam, S. S. . Keystone: A group key management service, International Conference on Telecommunications; 2000; p 2000. [Google Scholar]
  27. Hanchett, E. ; Listwon, B. . Vue. js in Action.; Simon and Schuster: 2018. [Google Scholar]
  28. Chakrabarti, S. ; Saito, N. ; Lyu, M. R. . et al. Integrating the document object model with hyperlinks for enhanced topic distillation and information extraction, Proceedings of the 10th international conference on World Wide Web; 2001; pp 211–220. [Google Scholar]
  29. Sheikh, W. ; Sheikh, N. . A Model-View-ViewModel (MVVM) application framework for hearing impairment diagnosis-class inheritance architecture, Intermountain Engineering, Technology and Computing; 2020; pp 1–6. [Google Scholar]
  30. Li Z., Wu Z., Mo G., Xing X., Liu P.. A small-angle X-ray scattering station at Beijing synchrotron radiation facility. Instrum. Sci. Technol. 2014;42(2):128–141. doi: 10.1080/10739149.2013.845845. [DOI] [Google Scholar]
  31. Hastings J. B., Thomlinson W., Cox D.. Synchrotron X-ray powder diffraction. Appl. Crystallogr. 1984;17(2):85–95. doi: 10.1107/S0021889884011043. [DOI] [Google Scholar]
  32. Chu B., Hsiao B. S.. Small-angle X-ray scattering of polymers. Chem. Rev. 2001;101(6):1727–1762. doi: 10.1021/cr9900376. [DOI] [PubMed] [Google Scholar]
  33. Fenter P., Sturchio N. C.. Mineral–water interfacial structures revealed by synchrotron X-ray scattering. Prog. Surf. Sci. 2004;77(5–8):171–258. doi: 10.1016/j.progsurf.2004.12.001. [DOI] [Google Scholar]
  34. Hendrickson W. A.. Determination of macromolecular structures from anomalous diffraction of synchrotron radiation. Science. 1991;254(5028):51–58. doi: 10.1126/science.1925561. [DOI] [PubMed] [Google Scholar]
  35. Sanchez-Cano C., Alvarez-Puebla R. A., Abendroth J. M., Beck T., Blick R., Cao Y., Caruso F., Chakraborty I., Chapman H. N., Chen C.. et al. X-ray-based techniques to study the nano–bio interface. ACS Nano. 2021;15(3):3754–3807. doi: 10.1021/acsnano.0c09563. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Dolabella S., Borzì A., Dommann A., Neels A.. Lattice strain and defects analysis in nanostructured semiconductor materials and devices by high-resolution X-ray diffraction: Theoretical and practical aspects. Small Methods. 2022;6(2):2100932. doi: 10.1002/smtd.202100932. [DOI] [PubMed] [Google Scholar]
  37. Thiele J. C., Pfitzner E., Kukura P.. Single-protein optical holography. Nat. Photonics. 2024;18(4):388–395. doi: 10.1038/s41566-024-01405-2. [DOI] [Google Scholar]
  38. Han J., Feng J., Kang J., Chen J.-M., Du X.-Y., Ding S.-Y., Liang L., Wang W.. Fast growth of single-crystal covalent organic frameworks for laboratory x-ray diffraction. Science. 2024;383(6686):1014–1019. doi: 10.1126/science.adk8680. [DOI] [PubMed] [Google Scholar]
  39. Li Y.-W., Liu G.-F., Wu H.-J., Zhou P., Hong C.-X., Li N., Bian F.-G.. BL19U2: Small-angle X-ray scattering beamline for biological macromolecules in solution at SSRF. Nucl. Sci. Tech. 2020;31(12):117. doi: 10.1007/s41365-020-00825-3. [DOI] [Google Scholar]
  40. Ralston C. Y., Gupta S., Del Mundo J. T., Soe A. C., Russell B., Tyler J., Paul S., Kahan D., Kristensen L., Subramanian S.. et al. ALS-ENABLE: creating synergy and opportunity at the Advanced Light Source synchrotron structural biology beamlines. Synchrotron Radiation. 2025;32(4):1059–1067. doi: 10.1107/S1600577525004205. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Hua W.-Q., Yang C.-M., Zhou P., Tian F., Lin J.-Y., Wang Y.-Z., Li X.-Y., Miao X.-R., Hong C.-X., Huang Q.-S.. et al. Time-resolved ultra-small-angle X-ray scattering beamline (BL10U1) at SSRF. Nucl. Sci. Tech. 2024;35(2):36. doi: 10.1007/s41365-024-01389-2. [DOI] [Google Scholar]
  42. Gu Y.-L., Zheng X., Zhang X.-M., Zhao B., Yin G.-Z., Zhu D.-M., Sun Y.-H., Gao X.-Y., Li X.-L.. Surface diffraction beamline at the SSRF. Nucl. Sci. Tech. 2024;35(7):116. doi: 10.1007/s41365-024-01490-6. [DOI] [Google Scholar]
  43. Hammersley A. P.. FIT2D: a multi-purpose data reduction, analysis and visualization program. Appl. Crystallogr. 2016;49(2):646–652. doi: 10.1107/S1600576716000455. [DOI] [Google Scholar]
  44. Filik J., Ashton A. W., Chang P. C. Y., Chater P., Day S., Drakopoulos M., Gerring M., Hart M., Magdysyuk O., Michalik S.. et al. Processing two-dimensional X-ray diffraction and small-angle scattering data in DAWN 2. Appl. Crystallogr. 2017;50(3):959–966. doi: 10.1107/S1600576717004708. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Toby B. H., Von Dreele R. B.. GSAS-II: the genesis of a modern open-source all purpose crystallography software package. Appl. Crystallogr. 2013;46(2):544–549. doi: 10.1107/S0021889813003531. [DOI] [Google Scholar]
  46. Breßler I., Kohlbrecher J., Thünemann A. F.. SASfit: a tool for small-angle scattering data analysis using a library of analytical expressions. Appl. Crystallogr. 2015;48(5):1587–1598. doi: 10.1107/S1600576715016544. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Momma K., Izumi F.. VESTA: a three-dimensional visualization system for electronic and structural analysis. Appl. Crystallogr. 2008;41(3):653–658. doi: 10.1107/S0021889808012016. [DOI] [Google Scholar]
  48. McCusker L. B., Von Dreele R., Cox D., Louër D., Scardi P.. Rietveld refinement guidelines. Appl. Crystallogr. 1999;32(1):36–50. doi: 10.1107/S0021889898009856. [DOI] [Google Scholar]
  49. De Groot F.. High-resolution X-ray emission and X-ray absorption spectroscopy. Chem. Rev. 2001;101(6):1779–1808. doi: 10.1021/cr9900681. [DOI] [PubMed] [Google Scholar]
  50. Fang L., Seifert S., Winans R. E., Li T.. Understanding synthesis and structural variation of nanomaterials through in situ/operando XAS and SAXS. Small. 2022;18(19):2106017. doi: 10.1002/smll.202106017. [DOI] [PubMed] [Google Scholar]
  51. Tuerxun F., Otani S., Yamamoto K., Matsunaga T., Imai H., Mandai T., Watanabe T., Uchiyama T., Kanamura K., Uchimoto Y.. Phase transition behavior of MgMn2O4 spinel oxide cathode during magnesium ion insertion. Chem. Mater. 2021;33(3):1006–1012. doi: 10.1021/acs.chemmater.0c04194. [DOI] [Google Scholar]
  52. Rehr J. J., Kas J. J., Vila F. D., Prange M. P., Jorissen K.. Parameter-free calculations of X-ray spectra with FEFF9. Phys. Chem. Chem. Phys. 2010;12(21):5503–5513. doi: 10.1039/b926434e. [DOI] [PubMed] [Google Scholar]
  53. Benfatto M., Della Longa S., Pace E., Chillemi G., Padrin C., Natoli C. R., Sanna N.. MXAN: A new program for ab-initio structural quantitative analysis of XANES experiments. Comput. Phys. Commun. 2021;265(1):107992. doi: 10.1016/j.cpc.2021.107992. [DOI] [Google Scholar]
  54. Guda A. A., Guda S. A., Martini A., Kravtsova A., Algasov A., Bugaev A., Kubrin S. P., Guda L., Šot P., van Bokhoven J. A.. et al. Understanding X-ray absorption spectra by means of descriptors and machine learning algorithms. npj Comput. Mater. 2021;7(1):203. doi: 10.1038/s41524-021-00664-9. [DOI] [Google Scholar]
  55. Wilhelm F., Rogalev A.. Advanced X-ray absorption spectroscopy under high pressures at the ESRF beamline ID12. High Pressure Res. 2024;44(3):294–309. doi: 10.1080/08957959.2024.2362745. [DOI] [Google Scholar]
  56. Chen Y., Gao Q., Jiang Z., Li J., Zhang S.. Quick-scanning X-ray absorption fine structure beamline at SSRF. Nucl. Sci. Tech. 2024;35(6):92. doi: 10.1007/s41365-024-01466-6. [DOI] [Google Scholar]
  57. Sakdinawat A., Attwood D.. Nanoscale X-ray imaging. Nat. Photonics. 2010;4(12):840–848. doi: 10.1038/nphoton.2010.267. [DOI] [Google Scholar]
  58. Li J., Gao Y., Kuang Y., Shi J., Du X., Zhou J., Wang H., Yang Z., Xu B.. Dephosphorylation of D-peptide derivatives to form biofunctional, supramolecular nanofibers/hydrogels and their potential applications for intracellular imaging and intratumoral chemotherapy. J. Am. Chem. Soc. 2013;135(26):9907–9914. doi: 10.1021/ja404215g. [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Gao Y., Shi J., Yuan D., Xu B.. Imaging enzyme-triggered self-assembly of small molecules inside live cells. Nat. Commun. 2012;3(1):1033. doi: 10.1038/ncomms2040. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Paleo P., Mirone A.. Ring artifacts correction in compressed sensing tomographic reconstruction. Synchrotron Radiation. 2015;22(5):1268–1278. doi: 10.1107/S1600577515010176. [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. Xi Z., Yao H., Zhang T., Su Z., Wang B., Feng W., Pu Q., Zhao L.. Quantitative Three-Dimensional Imaging Analysis of HfO2 Nanoparticles in Single Cells via Deep Learning Aided X-ray Nano-Computed Tomography. ACS Nano. 2024;18(33):22378–22389. doi: 10.1021/acsnano.4c06953. [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. Jia L.-H., Li R.-Y., Duan Z.-M., Jiang S.-D., Wang B.-W., Wang Z.-M., Gao S.. Hydrothermal synthesis, structures and magnetic studies of transition metal sulfates containing hydrazine. Inorg. Chem. 2011;50(1):144–154. doi: 10.1021/ic101488k. [DOI] [PubMed] [Google Scholar]
  63. Chykhrij S., Babizhetsky V., Kuz’ma Y. B.. New Ternary Phosphides Ln25Ni49P33 (Ln= Ce, Pr, Nd, Sm, Gd, Tb, Dy, Ho, Er) Z. Anorg. Allg. Chem. 2001;627(6):1319–1324. doi: 10.1002/1521-3749(200106)627:6<1319::AID-ZAAC1319>3.0.CO;2-U. [DOI] [Google Scholar]
  64. Shabani L., Kasaee S. R., Chelliapan S., Abbasi M., Khajehzadeh H., Dehghani F. S., Firuzyar T., Shafiee M., Amani A. M., Mosleh-Shirazi S.. et al. An investigation into green synthesis of Ru template gold nanoparticles and the in vitro photothermal effect on the MCF-7 human breast cancer cell line. Appl. Phys. A:Mater. Sci. Process. 2023;129(8):564. doi: 10.1007/s00339-023-06832-6. [DOI] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

jp6c01154_si_001.pdf (73.8KB, pdf)

Articles from The Journal of Physical Chemistry. a are provided here courtesy of American Chemical Society

RESOURCES