The Role of Structural Representation in the Performance of a Deep Neural Network for X-Ray Spectroscopy

Toward a Machine Learning Approach to Interpreting X-ray Spectra of Trace Impurities by Converting XANES to EXAFS

The fact that the photoabsorption spectrum of a material contains information about the atomic structure, commonly understood in terms of multiple scattering theory, is the basis of the popular extended X-ray absorption spectroscopy (EXAFS) technique. How much of the same structural information is present in other complementary spectroscopic signals is not obvious. Here we use a machine learning approach to demonstrate that within theoretical models that accurately predict the EXAFS signal, the extended near-edge region does indeed contain the EXAFS-accessible structural information. We do this by exhibiting deep operator neural networks (DeepONets) that have learned the relationship between the extended and near edge portions of the X-ray absorption spectrum to predict the former from the latter. We find that we can accurately predict the EXAFS spectrum between 6 and 14 Å-1 from the first 6 Å-1 (≈100 eV) of the absorption spectrum of Cu2+ substitutional defects in the Fe3+ mineral hematite (α-Fe2O3). This surprising finding implies that theoretical analyses of X-ray absorption spectra could be implemented that extract the same conclusions as high-quality EXAFS studies from spectra collected over a much smaller range of photon energies. This relaxes a host of experimental limitations related to the X-ray source and measurement sample, including collection time, minimum dopant concentration, source brilliance, and energy range. We describe the theoretical data sets and DeepONet construction and show that the resulting DeepONets produce EXAFS that recovers linear combination fits to experimental data with accuracy approaching the original ab initio calculations. We discuss the implications of our findings for minor constituent characterization and for understanding the information content of spectroscopic data more broadly, including how this approach might be applied to measured experimental spectra. To encourage similar efforts, the simulated X-ray spectra, machine learning, and fitting code are publicly available.

Partial density of states representation for accurate deep neural network predictions of X-ray spectra

The performance of a machine learning (ML) algorithm for chemistry is highly contingent upon the architect's choice of input representation. This work introduces the partial density of states (p-DOS) descriptor: a novel, quantum-inspired structural representation which encodes relevant electronic information for machine learning models seeking to simulate X-ray spectroscopy. p-DOS uses a minimal basis set in conjunction with a guess (non-optimised) electronic configuration to extract and then discretise the density of states (DOS) of the absorbing atom to form the input vector. We demonstrate that while the electronically-focused p-DOS performs well in isolation, optimal performance is achieved when supplemented with nuclear structural information imparted via a geometric representation. p-DOS provides a description of the key electronic properties of a system which is not only concise and computationally efficient, but also independent of molecular size or choice of basis set. It can be rapidly generated, facilitating its application with large training sets. Its performance is demonstrated using a wide variety of examples at the sulphur K-edge, including the prediction of ultrafast X-ray spectroscopic signal associated with photoexcited 2(5H)-thiophenone. These results highlight the potential for ML models developed using p-DOS to contribute to the interpretation and prediction of experimental results e.g. in operando measurements of batteries and/or catalysts and femtosecond time-resolved studies, especially those made possible by emergent cutting-edge technologies, especially X-ray free electron lasers.

Tracking nuclear motion in single-molecule magnets using femtosecond X-ray absorption spectroscopy

The development of new data storage solutions is crucial for emerging digital technologies. Recently, all-optical magnetic switching has been achieved in dielectrics, proving to be faster than traditional methods. Despite this, single-molecule magnets (SMMs), which are an important class of magnetic materials due to their nanometre size, remain underexplored for ultrafast photomagnetic switching. Herein, we report femtosecond time-resolved K-edge X-ray absorption spectroscopy (TR-XAS) on a Mn(III)-based trinuclear SMM. Exploiting the elemental specificity of XAS, we directly track nuclear dynamics around the metal ions and show that the ultrafast dynamics upon excitation of a crystal-field transition are dominated by a magnetically active Jahn-Teller mode. Our results, supported by simulations, reveal minute bond length changes from 0.01 to 0.05 Å demonstrating the sensitivity of the method. These geometrical changes are discussed in terms of magneto-structural relationships and consequently our results illustrate the importance of TR-XAS for the emerging area of ultrafast molecular magnetism.

An on-the-fly deep neural network for simulating time-resolved spectroscopy: predicting the ultrafast ring opening dynamics of 1,2-dithiane

Revolutionary developments in ultrafast light source technology are enabling experimental spectroscopists to probe the structural dynamics of molecules and materials on the femtosecond timescale. The capacity to investigate ultrafast processes afforded by these resources accordingly inspires theoreticians to carry out high-level simulations which facilitate the interpretation of the underlying dynamics probed during these ultrafast experiments. In this Article, we implement a deep neural network (DNN) to convert excited-state molecular dynamics simulations into time-resolved spectroscopic signals. Our DNN is trained on-the-fly from first-principles theoretical data obtained from a set of time-evolving molecular dynamics. The train-test process iterates for each time-step of the dynamics data until the network can predict spectra with sufficient accuracy to replace the computationally intensive quantum chemistry calculations required to produce them, at which point it simulates the time-resolved spectra for longer timescales. The potential of this approach is demonstrated by probing dynamics of the ring opening of 1,2-dithiane using sulphur K-edge X-ray absorption spectroscopy. The benefits of this strategy will be more markedly apparent for simulations of larger systems which will exhibit a more notable computational burden, making this approach applicable to the study of a diverse range of complex chemical dynamics.

Accurate, affordable, and generalizable machine learning simulations of transition metal x-ray absorption spectra using the XANESNET deep neural network

The affordable, accurate, and generalizable prediction of spectroscopic observables plays a key role in the analysis of increasingly complex experiments. In this article, we develop and deploy a deep neural network-XANESNET-for predicting the lineshape of first-row transition metal K-edge x-ray absorption near-edge structure (XANES) spectra. XANESNET predicts the spectral intensities using only information about the local coordination geometry of the transition metal complexes encoded in a feature vector of weighted atom-centered symmetry functions. We address in detail the calibration of the feature vector for the particularities of the problem at hand, and we explore the individual feature importance to reveal the physical insight that XANESNET obtains at the Fe K-edge. XANESNET relies on only a few judiciously selected features-radial information on the first and second coordination shells suffices along with angular information sufficient to separate satisfactorily key coordination geometries. The feature importance is found to reflect the XANES spectral window under consideration and is consistent with the expected underlying physics. We subsequently apply XANESNET at nine first-row transition metal (Ti-Zn) K-edges. It can be optimized in as little as a minute, predicts instantaneously, and provides K-edge XANES spectra with an average accuracy of ∼±2%-4% in which the positions of prominent peaks are matched with a >90% hit rate to sub-eV (∼0.8 eV) error.

Beyond structural insight: a deep neural network for the prediction of Pt L2/3-edge X-ray absorption spectra

X-ray absorption spectroscopy at the L_2/3 edge can be used to obtain detailed information about the local electronic and geometric structure of transition metal complexes. By virtue of the dipole selection rules, the transition metal L_2/3 edge usually exhibits two distinct spectral regions: (i) the "white line", which is dominated by bound electronic transitions from metal-centred 2p orbitals into unoccupied orbitals with d character; the intensity and shape of this band consequently reflects the d density of states (d-DOS), which is strongly modulated by mixing with ligand orbitals involved in chemical bonding, and (ii) the post-edge, where oscillations encode the local geometric structure around the X-ray absorption site. In this Article, we extend our recently-developed XANESNET deep neural network (DNN) beyond the K-edge to predict X-ray absorption spectra at the Pt L_2/3 edge. We demonstrate that XANESNET is able to predict Pt L_2/3 -edge X-ray absorption spectra, including both the parts containing electronic and geometric structural information. The performance of our DNN in practical situations is demonstrated by application to two Pt complexes, and by simulating the transient spectrum of a photoexcited dimeric Pt complex. Our discussion includes an analysis of the feature importance in our DNN which demonstrates the role of key features and assists with interpreting the performance of the network.

Zn-Induced Interactions Between SARS-CoV-2 orf7a and BST2/Tetherin

We present in this work a first X-ray Absorption Spectroscopy study of the interactions of Zn with human BST2/tetherin and SARS-CoV-2 orf7a proteins as well as with some of their complexes. The analysis of the XANES region of the measured spectra shows that Zn binds to BST2, as well as to orf7a, thus resulting in the formation of BST2-orf7a complexes. This structural information confirms the the conjecture, recently put forward by some of the present Authors, according to which the accessory orf7a (and possibly also orf8) viral protein are capable of interfering with the BST2 antiviral activity. Our explanation for this behavior is that, when BST2 gets in contact with Zn bound to the orf7a Cys15 ligand, it has the ability of displacing the metal owing to the creation of a new disulfide bridge across the two proteins. The formation of this BST2-orf7a complex destabilizes BST2 dimerization, thus impairing the antiviral activity of the latter.

Machine learning powered by principal component descriptors as the key for sorted structural fit of XANES

Modern synchrotron radiation sources and free electron laser made X-ray absorption spectroscopy (XAS) an analytical tool for the structural analysis of materials under in situ or operando conditions. Fourier approach applied to the extended region of the XAS spectrum (EXAFS) allows the estimation of the number of structural and non-structural parameters which can be refined through a fitting procedure. The near edge region of the XAS spectrum (XANES) is also sensitive to the coordinates of all the atoms in the local cluster around the absorbing atom. However, in contrast to EXAFS, the existing approaches of quantitative analysis provide no estimation for the number of structural parameters that can be evaluated for a given XANES spectrum. This problem exists both for the classical gradient descent approaches and for modern machine learning methods based on neural networks. We developed a new approach for rational fit based on principal component descriptors of the spectrum. In this work the principal component analysis (PCA) is applied to a dataset of theoretical spectra calculated a priori on a grid of variable structural parameters of a molecule or cluster. Each principal component of the dataset is related then to a combined variation of several structural parameters, similar to the vibrational normal mode. Orthogonal principal components determine orthogonal deformations that can be extracted independently upon the analysis of the XANES spectrum. Applying statistical criteria, the PCA-based fit of the XANES determines the accessible structural information in the spectrum for a given system.

Enhancing the analysis of disorder in X-ray absorption spectra: application of deep neural networks to T-jump-X-ray probe experiments

Many chemical and biological reactions, including ligand exchange processes, require thermal energy for the reactants to overcome a transition barrier and reach the product state. Temperature-jump (T-jump) spectroscopy uses a near-infrared (NIR) pulse to rapidly heat a sample, offering an approach for triggering these processes and directly accessing thermally-activated pathways. However, thermal activation inherently increases the disorder of the system under study and, as a consequence, can make quantitative interpretations of structural changes challenging. In this Article, we optimise a deep neural network (DNN) for the instantaneous prediction of Co K-edge X-ray absorption near-edge structure (XANES) spectra. We apply our DNN to analyse T-jump pump/X-ray probe data pertaining to the ligand exchange processes and solvation dynamics of Co2+ in chlorinated aqueous solution. Our analysis is greatly facilitated by machine learning, as our DNN is able to predict quickly and cost-effectively the XANES spectra of thousands of geometric configurations sampled from ab initio molecular dynamics (MD) using nothing more than the local geometric environment around the X-ray absorption site. We identify directly the structural changes following the T-jump, which are dominated by sample heating and a commensurate increase in the Debye-Waller factor.