Machine learning for analysis of experimental scattering and spectroscopy data in materials chemistry

Machine Learning Interpretation of Optical Spectroscopy Using Peak-Sensitive Logistic Regression

Optical spectroscopy, a noninvasive molecular sensing technique, offers valuable insights into material characterization, molecule identification, and biosample analysis. Despite the informativeness of high-dimensional optical spectra, their interpretation remains a challenge. Machine learning methods have gained prominence in spectral analyses, efficiently unveiling analyte compositions. However, these methods still face challenges in interpretability, particularly in generating clear feature importance maps that highlight the spectral features specific to each class of data. These limitations arise from feature noise, model complexity, and the lack of optimization for spectroscopy. In this work, we introduce a machine learning algorithm─logistic regression with peak-sensitive elastic-net regularization (PSE-LR)─tailored for spectral analysis. PSE-LR enables classification and interpretability by producing a peak-sensitive feature importance map, achieving an F1-score of 0.93 and a feature sensitivity of 1.0. Its performance is compared with other methods, including k-nearest neighbors (KNN), elastic-net logistic regression (E-LR), support vector machine (SVM), principal component analysis followed by linear discriminant analysis (PCA-LDA), XGBoost, and neural network (NN). Applying PSE-LR to Raman and photoluminescence (PL) spectra, we detected the receptor-binding domain (RBD) of SARS-CoV-2 spike protein in ultralow concentrations, identified neuroprotective solution (NPS) in brain samples, recognized WS2 monolayer and WSe2/WS2 heterobilayer, analyzed Alzheimer's disease (AD) brains, and suggested potential disease biomarkers. Our findings demonstrate PSE-LR's utility in detecting subtle spectral features and generating interpretable feature importance maps. It is beneficial for the spectral characterization of materials, molecules, and biosamples and applicable to other spectroscopic methods. This work also facilitates the development of nanodevices such as nanosensors and miniaturized spectrometers based on nanomaterials.

Surface-Enhanced Raman Spectroscopy for Biomedical Applications: Recent Advances and Future Challenges

The year 2024 marks the 50th anniversary of the discovery of surface-enhanced Raman spectroscopy (SERS). Over recent years, SERS has experienced rapid development and became a critical tool in biomedicine with its unparalleled sensitivity and molecular specificity. This review summarizes the advancements and challenges in SERS substrates, nanotags, instrumentation, and spectral analysis for biomedical applications. We highlight the key developments in colloidal and solid SERS substrates, with an emphasis on surface chemistry, hotspot design, and 3D hydrogel plasmonic architectures. Additionally, we introduce recent innovations in SERS nanotags, including those with interior gaps, orthogonal Raman reporters, and near-infrared-II-responsive properties, along with biomimetic coatings. Emerging technologies such as optical tweezers, plasmonic nanopores, and wearable sensors have expanded SERS capabilities for single-cell and single-molecule analysis. Advances in spectral analysis, including signal digitalization, denoising, and deep learning algorithms, have improved the quantification of complex biological data. Finally, this review discusses SERS biomedical applications in nucleic acid detection, protein characterization, metabolite analysis, single-cell monitoring, and in vivo deep Raman spectroscopy, emphasizing its potential for liquid biopsy, metabolic phenotyping, and extracellular vesicle diagnostics. The review concludes with a perspective on clinical translation of SERS, addressing commercialization potentials and the challenges in deep tissue in vivo sensing and imaging.

Nonlinear Harmonics: A Gateway to Enhanced Image Contrast and Material Discrimination

Recent advancements in atomic force microscopy (AFM) have enabled detailed exploration of materials at the molecular and atomic levels. These developments, however, pose a challenge: the data generated by microscopic and spectroscopic experiments are increasing rapidly in both size and complexity. Extracting meaningful physical insights from these datasets is challenging, particularly for multilayer heterogeneous nanoscale structures. In this paper, an unsupervised approach is presented to enhance AFM image contrast by analyzing the nonlinear response of a cantilever interacting with a material's surface using a wavelet-based AFM. This method simultaneously measures different frequencies and harmonics in a single scan, without the need for additional hardware and exciting multiple cantilevers' eigenmodes. This developed AFM image contrast enhancement (AFM-ICE) approach employs unsupervised learning, image processing, and image fusion techniques. The method is applied to interpret complex multilayer structures consist of defects, deposited nanoparticles and heterogeneities. Its substantial capability is demonstrated to improve image contrast and differentiate between various components. This methodology can pave the way for rapid and precise determination of material properties with enhanced resolution.

When can we trust structural models derived from pair distribution function measurements?

The pair distribution function (PDF) is an important metric for characterising structure in complex materials, but it is well known that meaningfully different structural models can sometimes give rise to equivalent PDFs. In this paper, we discuss the use of model likelihoods as a general approach for discriminating between such homometric structure solutions. Drawing on two main case studies-one concerning the structure of a small peptide and the other amorphous calcium carbonate-we show how consideration of model likelihood can help drive robust structure solution, even in cases where the PDF is particularly information-poor. The obvious thread of these individual case studies is the potential role for machine-learning approaches to help guide structure determination from the PDF, and our paper finishes with some forward-looking discussion along these lines.

A Vision for the Future of Neutron Scattering and Muon Spectroscopy in the 2050s

Neutron scattering and muon spectroscopy are techniques that use subatomic particles to understand materials across a wide range of energy (μeV to tens of eV), length (Å to cm) and time (attosecond to hour) scales. The methods are widely used to study condensed phase materials in areas that span physics, chemistry, biology, engineering and cultural heritage. In this Perspective we consider three questions: (i) will neutron scattering and muon spectroscopy still be needed in the 2050s? (ii) What might the technology to produce neutron and muon beams look like in the 2050s? (iii) What will be the applications in the 2050s? Overall, the neutron/muon ecosystem in the 2050s will have less capacity than now, but greater capability because of the somewhat higher power sources, better instrumentation and data analysis.

Predicting Colloidal Interaction Parameters from Small-Angle X-ray Scattering Curves Using Artificial Neural Networks and Markov Chain Monte Carlo Sampling

Small-angle X-ray scattering (SAXS) is a characterization technique that allows for the study of colloidal interactions by fitting the structure factor of the SAXS profile with a selected model and closure relation. However, the applicability of this approach is constrained by the limited number of existing models that can be fitted analytically, as well as the narrow operating range for which the models are valid. In this work, we demonstrate a proof of concept for using an artificial neural network (ANN) trained on SAXS curves obtained from Monte Carlo (MC) simulations to predict values of the effective macroion valency (Z eff) and the Debye length (κ-1) for a given SAXS profile. This ANN, which was trained on 200,000 simulated SAXS curves, was able to predict values of Z eff and κ-1 for a test set containing 25,000 simulated SAXS curves, where most predicted values had errors smaller than 20%. Subsequently, an ANN was used as a surrogate model in a Markov chain Monte Carlo sampling algorithm to obtain maximum a posteriori estimates of Z eff and κ-1, as well as the associated confidence intervals and correlations between Z eff and κ-1 for an experimentally obtained SAXS profile.

Hepatic toxicity prediction of bisphenol analogs by machine learning strategy

Toxicological studies have demonstrated the hepatic toxicity of several bisphenol analogs (BPs), a prevalent type of endocrine disruptor. The development of Adverse Outcome Pathway (AOP) has substantially contributed to the rapid risk assessment for human health. However, the lack of in vitro and in vivo data for the emerging BPs has limited the hazard assessment of these synthetic chemicals. Here, we aimed to develop a new strategy to rapidly predict BPs' hepatotoxicity using network analysis coupled with machine learning models. Considering the structural and functional similarities shared by BPs with Bisphenol A (BPA), we first integrated hepatic disease related genes from multiple databases into BPA-Gene-Phenotype-hepatic toxicity network and subjected it to the computational AOP (cAOP). Through cAOP network and conventional machine learning approaches, we scored the hepatotoxicity of 20 emerging BPs and provided new insights into how BPs' structure features contributed to biologic functions with limited experimental data. Additionally, we assessed the interactions between emerging BPs and ESR1 using molecular docking and proposed an AOP framework wherein ESR1 was a molecular initiating event. Overall, our study provides a computational approach to predict the hepatotoxicity of emerging BPs.

Gaining More Insights from Synchrotron-Based X-ray Spectroscopy for Alkali Ion Rechargeable Batteries

Alkali ion rechargeable batteries play a significant part in portable electronic devices and electronic vehicles. The rapid development of renewable energy technology nowadays demands batteries with even higher energy density for grid storage. To fulfill such demand, extensive research efforts have been devoted to optimizing electrochemical properties as well as developing novel energy storage schemes and designing new systems. In the investigation process, synchrotron-based X-ray spectroscopy plays a vital role in investigating the detailed degradation mechanism and developing novel energy storage schemes. Herein, we critically review the applications of synchrotron-based X-ray spectroscopy in battery research in recent years. This review begins with a discussion of the different scientific issues in alkali ion rechargeable batteries within various time and space scales. Subsequently, the principle of synchrotron-based X-ray spectroscopy is introduced, and the characteristics of various characterization techniques are summarized and compared. Typical application cases of synchrotron-based X-ray spectroscopy are then introduced into battery investigations. The final part presents perspectives in the development direction of both alkali ion rechargeable battery systems and synchrotron-based X-ray spectroscopy in the future.

Computational Reverse Engineering Analysis of the Scattering Experiment Method for Interpretation of 2D Small-Angle Scattering Profiles (CREASE-2D)

Small-angle scattering (SAS) is a widely used characterization technique that provides structural information in soft materials at varying length scales (nanometers to microns). The output of an SAS measurement is the scattered intensity I(q) as a function of q, the scattered wavevector with respect to the incident wave; the latter is represented by its magnitude |q| ≡ q (in inverse distance units) and azimuthal angle θ. While isotropic structural arrangement can be interpreted by analysis of the azimuthally averaged one-dimensional (1D) scattering profile, to understand anisotropic arrangements, one has to interpret the two-dimensional (2D) scattering profile, I(q, θ). Manual interpretation of such 2D profiles usually involves fitting of approximate analytical models to azimuthally averaged sections of the 2D profile. In this paper, we present a new method called CREASE-2D that interprets, without any azimuthal averaging, the entire 2D scattering profile, I(q, θ), and outputs the relevant structural features. CREASE-2D is an extension of the "computational reverse engineering analysis for scattering experiments" (CREASE) method that has been used successfully to analyze 1D SAS profiles for a variety of soft materials. CREASE-2D goes beyond CREASE by enabling analysis of 2D scattering profiles, which is far more challenging to interpret than the azimuthally averaged 1D profiles. The CREASE-2D workflow identifies the structural features whose computed I(q, θ) profiles, calculated using a surrogate XGBoost machine learning model, match the input experimental I(q, θ). We expect that this CREASE-2D method will be a valuable tool for materials' researchers who need direct interpretation of the 2D scattering profiles in contrast to analyzing azimuthally averaged 1D I(q) vs q profiles that can lose important information related to structural anisotropy.

Enhancing antioxidant properties of CeO2 nanoparticles with Nd3+ doping: structural, biological, and machine learning insights

The antioxidant capabilities of nanoparticles are contingent upon various factors, including their shape, size, and chemical composition. Herein, novel Nd-doped CeO2 nanoparticles were synthesized and the neodymium content was varied to investigate the synergistic impact on the antioxidant properties of CeO2 nanoparticles. Incorporating Nd3+ induced changes in lattice parameters and significantly altered the morphology from nanoparticles to nanorods. The biological activity of Nd-doped CeO2 was examined against pathogenic bacterial strains, breast cancer cell lines, and antioxidant models. The antibacterial and anticancer activities of nanoparticles were not observed, which could be associated with the Ce3+/Ce4+ ratio. Notably, the incorporation of neodymium improved the antioxidant capacity of CeO2. Machine learning techniques were employed to forecast the antioxidant activity to enhance understanding and predictive capabilities. Among these models, the random forest model exhibited the highest accuracy at 96.35%, establishing it as a robust computational tool for elucidating the biological behavior of Nd-doped CeO2 nanoparticles. This study presents the first exploration of the influence of Nd3+ on the structural, optical, and biological attributes of CeO2, contributing valuable insights and extending the application of machine learning in predicting the therapeutic efficacy of inorganic nanomaterials.

POMFinder: identifying polyoxometallate cluster structures from pair distribution function data using explainable machine learning

Characterization of a material structure with pair distribution function (PDF) analysis typically involves refining a structure model against an experimental data set, but finding or constructing a suitable atomic model for PDF modelling can be an extremely labour-intensive task, requiring carefully browsing through large numbers of possible models. Presented here is POMFinder, a machine learning (ML) classifier that rapidly screens a database of structures, here polyoxometallate (POM) clusters, to identify candidate structures for PDF data modelling. The approach is shown to identify suitable POMs from experimental data, including in situ data collected with fast acquisition times. This automated approach has significant potential for identifying suitable models for structure refinement to extract quantitative structural parameters in materials chemistry research. POMFinder is open source and user friendly, making it accessible to those without prior ML knowledge. It is also demonstrated that POMFinder offers a promising modelling framework for combined modelling of multiple scattering techniques.