Search NASA⌕ Search

SEARCH · Search NASA

Results for “Convolution”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,261 records · Page 70

Automatic detection of cataclysmic variables from SDSS images

Abstract Investigating rare and new objects have always been an important direction in astronomy. Cataclysmic variables (CVs) are ideal and natural celestial bodies for studying the accretion process of semi-detached binaries with accretion processes. However, the sample size of CVs must increase because a lager gap exists between the observational and the theoretical expanding CVs. Astronomy has entered the big data era and can provide massive images containing CV candidates. CVs as a type of faint celestial objects, are highly challenging to be identified directly from images using automatic manners. Deep learning has rapidly developed in intelligent image processing and has been widely applied in some astronomical fields with excellent detection results. YOLOX, as the latest YOLO framework, is advantageous in detecting small and dark targets. This work proposes an improved YOLOX-based framework according to the characteristics of CVs and Sloan Digital Sky Survey (SDSS) photometric images to train and verify the model to realise CV detection. We use the Convolutional Block Attention Module to increase the number of output features with the feature extraction network and adjust the feature fusion network to obtain fused features. Accordingly, the loss function is modified. Experimental results demonstrate that the improved model produces satisfactory results, with average accuracy (mean average Precision at 0.5) of 92.0%, Precision of 92.9%, Recall of 94.3%, and $F1-score$ of 93.6% on the test set. The proposed method can efficiently achieve the identification of CVs in test samples and search for CV candidates in unlabeled images. The image data vastly outnumber the spectra in the SDSS-released data. With supplementary follow-up observations or spectra, the proposed model can help astronomers in seeking and detecting CVs in a new manner to ensure that a more extensive CV catalog can be built. The proposed model may also be applied to the detection of other kinds of celestial objects.

Astronomy & Astrophysics↗

Statistical characterization of experimental magnetized liner inertial fusion stagnation images using deep-learning-based fuel–background segmentation

Significant variety is observed in spherical crystal x-ray imager (SCXI) data for the stagnated fuel–liner system created in Magnetized Liner Inertial Fusion (MagLIF) experiments conducted at the Sandia National Laboratories Z-facility. As a result, image analysis tasks involving, e.g., region-of-interest selection (i.e. segmentation), background subtraction and image registration have generally required tedious manual treatment leading to increased risk of irreproducibility, lack of uncertainty quantification and smaller-scale studies using only a fraction of available data. We present a convolutional neural network (CNN)-based pipeline to automate much of the image processing workflow. This tool enabled batch preprocessing of an ensemble of N scans = 139 SCXI images across N exp = 67 different experiments for subsequent study. The pipeline begins by segmenting images into the stagnated fuel and background using a CNN trained on synthetic images generated from a geometric model of a physical three-dimensional plasma. The resulting segmentation allows for a rules-based registration. Our approach flexibly handles rarely occurring artifacts through minimal user input and avoids the need for extensive hand labelling and augmentation of our experimental dataset that would be needed to train an end-to-end pipeline. Here we also fit background pixels using low-degree polynomials, and perform a statistical assessment of the background and noise properties over the entire image database. Our results provide a guide for choices made in statistical inference models using stagnation image data and can be applied in the generation of synthetic datasets with realistic choices of noise statistics and background models used for machine learning tasks in MagLIF data analysis. We anticipate that the method may be readily extended to automate other MagLIF stagnation imaging applications.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

How does ion temperature gradient turbulence depend on magnetic geometry? Insights from data and machine learning

Magnetic geometry has a significant effect on the level of turbulent transport in fusion plasmas. Here, we model and analyse this dependence using multiple machine learning methods and a dataset of >200 000 nonlinear gyrokinetic simulations of ion-temperature-gradient turbulence in diverse non-axisymmetric geometries. The dataset is generated using a large collection of both optimised and randomly generated stellarator equilibria. At fixed gradients and other input parameters, the turbulent heat flux varies between geometries by several orders of magnitude. Trends are apparent among the configurations with particularly high or particularly low heat flux. Regression and classification techniques from machine learning are then applied to extract patterns in the dataset. Due to a symmetry of the gyrokinetic equation, the heat flux and regressions thereof should be invariant to translations of the raw features in the parallel coordinate, similar to translation invariance in computer vision applications. Multiple regression models including convolutional neural networks (CNNs) and decision trees can achieve reasonable predictive power for the heat flux in held-out test configurations, with highest accuracy for the CNNs. Using Spearman correlation, sequential feature selection and Shapley values to measure feature importance, it is consistently found that the most important geometric lever on the heat flux is the flux surface compression in regions of bad curvature. The second most important geometric feature relates to the magnitude of geodesic curvature. These two features align remarkably with surrogates that have been proposed based on theory, while the methods here allow a natural extension to more features for increased accuracy. The dataset, released with this publication, may also be used to test other proposed surrogates, and we find that many previously published proxies do correlate well with both the heat flux and stability boundary.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Separation of Protein Conformers by Differential Ion Mobility in Hydrogen-Rich Gases

Proteins in solution or the gas phase tend to exhibit multiple conformational families, each comprising distinct structures. Separation methods have generally failed to resolve these, with their convolution producing wide peaks. Here we report full separation of >10 conformers for most ubiquitin charge states by the new approach of differential ion mobility spectrometry (FAIMS) employing H 2 /N 2 gas mixtures with up to 85% H 2 . The resolving power (up to 400) is five times the highest previously achieved (using He/N 2 buffers), greatly increasing the separation specificity. The peak widths match the narrowest obtained by FAIMS for any species under same conditions and scale with the protein charge state (z) and ion residence time (t) as z-1/2 and t-1/2, as prescribed for instrumental (diffusional) broadening. This suggests resolution of specific geometries rather than broader ensembles.

08 HYDROGEN↗

GlyQ-IQ: Glycomics Quintavariate-Informed Quantification with High-Performance Computing and GlycoGrid 4D Visualization

Dense LC-MS datasets have convoluted extracted ion chromatograms with multiple chromatographic peaks that cloud the differentiation between intact compounds with their overlapping isotopic distributions, peaks due to insource ion fragmentation, and noise. Making this differentiation is critical in glycomics datasets because chromatographic peaks correspond to different intact glycan structural isomers. The GlyQ-IQ software is targeted chromatography centric software designed for chromatogram and mass spectra data processing and subsequent glycan composition annotation. The targeted analysis approach offers several key advantages to LC-MS data processing and annotation over traditional algorithms. A priori information about the individual target’s elemental composition allows for exact isotope profile modeling for improved feature detection and increased sensitivity by focusing chromatogram generation and peak fitting on the isotopic species in the distribution having the highest intensity and data quality. Glycan target annotation is corroborated by glycan family relationships and in source fragmentation detection. The GlyQ-IQ software is developed in this work (Part 1) and was used to profile N-glycan compositions from human serum LC-MS Datasets. The companion manuscript GlyQ-IQ Part 2 discusses developments in human serum N-glycan sample preparation, glycan isomer separation, and glycan electrospray ionization. A case study is presented to demonstrate how GlyQ-IQ identifies and removes confounding chromatographic peaks from high mannose glycan isomers from human blood serum. In addition, GlyQ-IQ was used to generate a broad N-glycan profile from a high resolution (100K/60K) nESI-LS-MS/MS dataset including CID and HCD fragmentation acquired on a Velos Pro Mass spectrometer. 101 glycan compositions and 353 isomer peaks were detected from a single sample. 99% of the GlyQ-IQ glycan-feature assignments passed manual validation and are backed with high resolution mass spectra and mass accuracies less than 7 ppm.

Environmental Molecular Sciences Laboratory↗

Coherence Spectroscopy in the Condensed Phase: Insights into Molecular Structure, Environment, and Interactions

The role of coherences, or coherently excited superposition states, in complex condensed-phase systems has been the topic of intense interest and debate for a number of years. In many cases, coherences have been utilized as spectators of ultrafast dynamics or for identifying couplings between electronic states. In rare cases, they have been found to drive excited state dynamics directly. Interestingly though, the utilization of coherences as a tool for high-detail vibronic spectroscopy has largely been overlooked until recently, despite their encoding of key information regarding molecular structure, electronically sensitive vibrational modes, and intermolecular interactions. Furthermore, their detection in the time domain makes for a highly comprehensive spectroscopic technique wherein the phase and dephasing times are extracted in addition to amplitude and intensity, an element not afforded in analogous frequency domain “steady-state” measurements. However, practical limitations arise in disentangling the large number of coherent signals typically accessed in broadband nonlinear spectroscopic experiments, often complicating assignment of the origin and type of coherences generated. Two-dimensional electronic spectroscopy (2DES) affords an avenue by which to disperse and decompose the large number of coherent signals generated in nonlinear experiments, facilitating the assignment of various types of quantum coherences. 2DES takes advantage of the broad bandwidth necessary for achieving the high time resolution desired for ultrafast dynamics and coherence generation by resolving the excitation axis to detect all excitation channels independently. This feature is beneficial for following population dynamics such as electronic energy transfer, and 2DES has become the choice method for such studies. Simultaneously, coherences arise as oscillations at well-defined coordinates across the 2D map often atop those evolving population signals. By isolating the coherent contribution to the 2DES data and Fourier transforming along the population time, a 3D spectral representation of the coherent 2D data is generated, and coherences are then ordered by their oscillation frequency, ν2. Individual coherences can then be selected by their frequency and evaluated via their distinct “2D coherence” spectra, yielding a significantly more distinctive set of spectroscopic signatures over other 1D methodologies and single-point 2DES analysis. Given that coherences of different origin result in unique 2D coherence spectra, these characteristics can be catalogued and compared directly against experiment for prompt assignment, a strategy not afforded by traditional 2DES analysis. In this Account, a structure-driven time-independent spectral model is discussed and employed to compare the 2D fingerprints of various coherences to experimental 2D coherence spectra. The frequency-domain approach can easily integrate ab initio derived vibronic parameters, and its correspondence with experimental coherence spectra of a model compound is demonstrated. Several examples and applications are discussed herein, from 2D Franck–Condon analysis of a model compound, to identifying the signatures of interpigment vibronic coupling in a photosynthetic light-harvesting complex. The 3D spectral approach to 2DES provides remarkable spectroscopic detail, in turn leading to new insights in molecular structure and interactions, which complement the time-resolved dynamics simultaneously recorded. The approach presented herein has the potential to distill down the convoluted set of nonlinear signals appearing in 2D coherent spectra, making the technique more amenable to high-detail vibronic spectroscopy in inherently complex condensed phase systems.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Anion Exchange in Cationic Frameworks: Structures of Channel-Forming Triarylpyrylium Tetrafluoroborate Salts

The crystal structure of a co-crystal of 2,4,6-tris(4-methylphenyl)pyridine and 2,4,6-tris(4-methylphenyl)pyrylium revealed a pseudohexagonal motif assembled from π-stacked helices that enclose 1-D channels containing disordered tetrafluoroborate counterions and solvent molecules. Though convoluted by disorder, aperiodicity, and complex twinning patterns, further investigation was motivated by the unique packing arrangement of this material, which enables solid-state anion diffusion, while resisting dissolution in water. Such properties are ideal for the development of anion exchange materials that can be employed for the remediation of agricultural and nuclear waste. Here, to develop a more complete understanding of the intermolecular interactions that give rise to this helical motif, we have compiled and analyzed the local interactions and extended structural motifs of 19 pore- and channel-forming 2,4,6-triarylpyrylium crystal structures. Here, three conserved local interactions were identified, including C–H···X– interactions, the rhombic binding pocket, and four-ring antiparallel dimerization. Development of this library of interaction motifs rationalizes the packing arrangement of the co-crystal and contributes to the informed design of anion exchange materials.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Probabilistic Deep Learning Approach to Automate the Interpretation of Multi-phase Diffraction Spectra

Autonomous synthesis and characterization of inorganic materials requires the automatic and accurate analysis of X-ray diffraction spectra. For this task, we designed a probabilistic deep learning algorithm to identify complex multi-phase mixtures. At the core of this algorithm lies an ensemble convolutional neural network trained on simulated diffraction spectra, which are systematically augmented with physics-informed perturbations to account for artifacts that can arise during experimental sample preparation and synthesis. In this work, larger perturbations associated with off-stoichiometry are also captured by supplementing the training set with hypothetical solid solutions. Spectra containing mixtures of materials are analyzed with a newly developed branching algorithm that utilizes the probabilistic nature of the neural network to explore suspected mixtures and identify the set of phases that maximize confidence in the prediction. Our model is benchmarked on simulated and experimentally measured diffraction spectra, showing exceptional performance with accuracies exceeding those given by previously reported methods based on profile matching and deep learning. We envision that the algorithm presented here may be integrated in experimental workflows to facilitate the high-throughput and autonomous discovery of inorganic materials.

36 MATERIALS SCIENCE↗

Synthesis and characterization of (Sn,Zn)O alloys

SnO exhibits electrical properties that render it promising for solar energy conversion applications, but it also has a strongly indirect band gap. Recent theoretical calculations predict that this disadvantage can be mitigated by isovalent alloying with other group-II oxides such as ZnO. Here, we synthesized new metastable isovalent (Sn,Zn)O alloy thin films by combinatorial reactive co-sputtering and characterized their structural, optical and electrical properties. The alloying of ZnO into SnO leads to a change of the valence state of the tin from Sn 0 via Sn 2+ to Sn 4+ , which can be counteracted by reducing the oxygen partial pressure during the deposition. The optical characterization of the smooth <10 at. % Sn 1-x Zn x O thin films showed an increase in the absorption coefficient in the range from 1 to 2 eV, which is consistent with the theoretical predictions for the isovalent alloying. However, the experimentally observed alloying effect may be convoluted with the effect of local variations of the Sn oxidation state. As a result, this effect would have to be minimized to improve the (Sn,Zn)O optical and electrical properties for their use as absorbers in solar energy conversion applications.

14 SOLAR ENERGY↗

Prospects and Limitations of Predicting Fuel Ignition Properties from Low-Temperature Speciation Data

Using chemical kinetic modeling and statistical analysis, we investigate the possibility of correlating key chemical “markers”–typically small molecules–formed during very lean (φ ~ 0.001) oxidation experiments with near-stoichiometric (φ ~ 1) fuel ignition properties. One goal of this work is to evaluate the feasibility of designing a fuel-screening platform, based on small laboratory reactors that operate at low temperatures and use minimal fuel volume. Buras et al. [Combust. Flame2020,216, 472–484] have shown that convolutional neural net (CNN) fitting can be used to correlate first-stage ignition delay times (IDTs) with OH/HO2 measurements during very lean oxidation in low-T flow reactors with better than factor-of-2 accuracy. In this work, we test the limits of applying this correlation-based approach to predict the low-temperature heat release (LTHR) and total IDT, including the sensitivity of total IDT to the equivalence ratio, φ. We demonstrate that first-stage IDT can be reliably correlated with very lean oxidation measurements using compressed sensing (CS), which is simpler to implement than CNN fitting. LTHR can also be predicted via CS analysis, although the correlation quality is somewhat lower than for first-stage IDT. In contrast, the accuracy of total IDT prediction at φ = 1 is significantly lower (within a factor of 4 or worse). Furthermore, these results can be rationalized by the fact that the first-stage IDT and LTHR are primarily determined by low-temperature chemistry, whereas total IDT depends on low-, intermediate-, and high-temperature chemistry. Oxidation reactions are most important at low temperatures, and therefore, measurements of universal molecular markers of oxidation do not capture the full chemical complexity required to accurately predict the total IDT even at a single equivalence ratio. As a result, we find that φ-sensitivity of ignition delay cannot be predicted at all using solely correlation with lean low-T chemical speciation measurements.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Structural Complexity and Tuned Thermoelectric Properties of a Polymorph of the Zintl Phase Ca 2 CdSb 2 with a Non-centrosymmetric Monoclinic Structure

The Zintl phase Ca 2 CdSb 2 was found to be dimorphic. Besides the orthorhombic Ca 2 CdSb 2 (-o), here we report on the synthesis, the structural characterization, and the thermoelectric transport properties of its monoclinic form, Ca 2 CdSb 2 (-m), and its Lu-doped variant Ca 2–x Lu x CdSb 2 (x ≈ 0.02). The monoclinic structure exhibits complex structural characteristics and constitutes a new structure type with the non-centrosymmetric space group Cm (Z = 30). The electrical resistivity ρ(T) measured on single crystals of both phases portrays a transition from a semiconductor to a degenerate p-type semiconductor upon doping with Lu and with an attendant change in the Hall carrier concentration nH from 7.15 × 10 18 to 2.30 × 10 19 cm –3 at 300 K. The Seebeck coefficient S(T) of both phases are comparable and indicate a hole-dominated carrier transport mechanism with magnitudes of 133 and 116 μV/K at 600 K for Ca 2 CdSb 2 (-m) and Ca 2–x Lu x CdSb 2 , respectively. The convoluted atomic bonding with an attendant large unit cell volume of ~4365 Å 3 drives a putative low thermal conductivity in these materials resulting in a power factor PF of 1.63 μW/cm K 2 and an estimated thermoelectric figure of merit zT of ~0.5 for Ca 2–x Lu x CdSb 2 at 600 K. Furthermore, differential scanning calorimetry results reveal the stability of these phases up to about 960 K, making them candidates for moderate temperature thermoelectric materials.

36 MATERIALS SCIENCE↗

Search for Stable and Low-Energy Ce–Co–Cu Ternary Compounds Using Machine Learning

Cerium-based intermetallics have garnered significant research attention as potential new permanent magnets. In this study, we explore the compositional and structural landscape of Ce−Co−Cu ternary compounds using a machine learning (ML)- guided framework integrated with first-principles calculations. We employ a crystal graph convolutional neural network (CGCNN), which enables efficient screening for promising candidates, significantly accelerating the material discovery process. With this approach, we predict five stable compounds, Ce 3 Co 3 Cu, CeCoCu 2 , Ce 12 Co 7 Cu, Ce 11 Co 9 Cu, and Ce 10 Co 11 Cu 4 , with formation energies below the convex hull, along with hundreds of low-energy (possibly metastable) Ce−Co−Cu ternary compounds. Firstprinciples calculations reveal that several structures are both energetically and dynamically stable. Notably, two Co-rich low-energy compounds, Ce 4 Co 33 Cu and Ce 4 Co 31 Cu 3 , are predicted to have high magnetizations.

Chemical structure↗

Predicting Small Molecule Transfer Free Energies by Combining Molecular Dynamics Simulations and Deep Learning

Accurately predicting small molecule partitioning and hydrophobicity is critical in the drug discovery process. There are many heterogeneous chemical environments within a cell and entire human body. For example, drugs must be able to cross the hydrophobic cellular membrane to reach their intracellular targets, and hydrophobicity is an important driving force for drug–protein binding. Atomistic molecular dynamics (MD) simulations are routinely used to calculate free energies of small molecules binding to proteins, crossing lipid membranes, and solvation but are computationally expensive. Machine learning (ML) and empirical methods are also used throughout drug discovery but rely on experimental data, limiting the domain of applicability. We present atomistic MD simulations calculating 15,000 small molecule free energies of transfer from water to cyclohexane. This large data set is used to train ML models that predict the free energies of transfer. We show that a spatial graph neural network model achieves the highest accuracy, followed closely by a 3D-convolutional neural network, and shallow learning based on the chemical fingerprint is significantly less accurate. A mean absolute error of ~4 kJ/mol compared to the MD calculations was achieved for our best ML model. We also show that including data from the MD simulation improves the predictions, tests the transferability of each model to a diverse set of molecules, and show multitask learning improves the predictions. This work provides insight into the hydrophobicity of small molecules and ML cheminformatics modeling, and our data set will be useful for designing and testing future ML cheminformatics methods.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Dynamic Profiling of β-Coronavirus 3CL Mpro Protease Ligand-Binding Sites

β-coronavirus (CoVs) alone has been responsible for three major global outbreaks in the 21st century. The current crisis has led to an urgent requirement to develop therapeutics. Even though a number of vaccines are available, alternative strategies targeting essential viral components are required as a backup against the emergence of lethal viral variants. One such target is the main protease (Mpro) that plays an indispensable role in viral replication. The availability of over 270 Mpro X-ray structures in complex with inhibitors provides unique insights into ligand–protein interactions. Herein, we provide a comprehensive comparison of all nonredundant ligand-binding sites available for SARS-CoV2, SARS-CoV, and MERS-CoV Mpro. Extensive adaptive sampling has been used to investigate structural conservation of ligand-binding sites using Markov state models (MSMs) and compare conformational dynamics employing convolutional variational auto-encoder-based deep learning. Our results indicate that not all ligand-binding sites are dynamically conserved despite high sequence and structural conservation across β-CoV homologs. This highlights the complexity in targeting all three Mpro enzymes with a single pan inhibitor.

59 BASIC BIOLOGICAL SCIENCES↗

A Multiple Filter Based Neural Network Approach to the Extrapolation of Adsorption Energies on Metal Surfaces for Catalysis Applications

Computational catalyst discovery involves the development of microkinetic reactor models based on estimated parameters determined from density functional theory (DFT). For complex surface chemistries, the number of reaction intermediates can be very large, and the cost of calculating the adsorption energies by DFT for all surface intermediates even for one active site model can become prohibitive. In this paper, we have identified appropriate descriptors and machine learning models that can be used to predict a significant part of these adsorption energies given data on the rest of them. Furthermore, our investigations also included the case when the species data used to train the predictive model are of different size relative to the species the model tries to predict—this is an extrapolation in the data space which is typically difficult with regular machine learning models. Due to the relative size of the available data sets, we have attempted to extrapolate from the larger species to the smaller ones in the current work. Here, we have developed a neural network based predictive model that combines an established additive atomic contribution based model with the concepts of a convolutional neural network that, when extrapolating, achieves a statistically significant improvement over the previous models.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Modeling the Conformer-Dependent Electronic Absorption Spectra and Photolysis Rates of Methyl Vinyl Ketone Oxide and Methacrolein Oxide

Criegee intermediates are important atmospheric oxidants, formed via the reaction of ozone with volatile alkenes emitted into the troposphere. Small Criegee intermediates (e.g., CH 2 OO and CH 3 CHOO) are highly reactive, and their removal via unimolecular decay or bimolecular chemistry dominates their atmospheric lifetimes. As the molecular complexity of Criegee intermediates increases, their electronic absorption spectra show a bathochromic shift within the solar spectrum relevant to the troposphere. In such cases, solar photolysis may become a competitive contributor to their atmospheric removal. In this article, we report the conformer-dependent simulated electronic absorption spectra of two four-carbon-centered Criegee intermediates, methyl vinyl ketone oxide (MVK-oxide) and methacrolein oxide (MACR-oxide). Both MVK-oxide and MACR-oxide contain four low-energy conformers, which are convoluted in the experimentally measured spectra. Here, we deconvolute each conformer and estimate contributions from each of the four conformers to the experimentally measured spectra. We also estimate the photolysis rates and predict that solar photolysis should be a more competitive removal process for MVK-oxide and MACR-oxide (cf. CH 2 OO and CH 3 CHOO).

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Comparison of Machine Learning Approaches for Prediction of the Equivalent Alkane Carbon Number for Microemulsions Based on Molecular Properties

The chemical properties of oils are vital in the design of microemulsion systems. The hydrophilic–lipophilic difference equation used to predict microemulsions’ phase behavior expresses the oils’ physiochemical properties as the equivalent alkane carbon number (EACN). The experimental determination of EACN requires knowledge of the temperature dependence of the microemulsion system and the effects of different surfactant concentrations. Thus, the experimental determination is time-intensive and tedious, requiring days to months for proper separations. Furthermore, the experiments require high purity of chemicals because microemulsions are sensitive to impurities. Our work focuses on the quick and reliable predictions of the EACN with machine learning (ML) models. Due to the immaturity of ML chemical predictions, we compare three graph neural networks (GNNs) and a gradient-boosted tree algorithm, known as XGBoost. The GNNs use the molecular structures represented as simplified molecular-input line-entry system (SMILES) codes for the initial input, which allows us to assess whether geometry optimization is necessary for reliable results. The XGBoost model also begins with the SMILES representations of the molecules but uses molecular descriptors instead of geometry optimizations. As a result, the best model tested (crystal graph convolutional neural network with Merck molecular force field-94) has an error of 1.15 EACN units of the true EACN for unknown data with the errors skewed toward zero and an R² score of 0.9

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Self-Evolving Machine: A Continuously Improving Model for Molecular Thermochemistry

Because collecting precise and accurate chemistry data is often challenging, chemistry data sets usually only span a small region of chemical space, which limits the performance and the scope of applicability of data-driven models. To address this issue, we integrated an active learning machine with automatic ab initio calculations to form a self-evolving model that can continuously adapt to new species appointed by the users. In the present work, we demonstrate the self-evolving concept by modeling the formation enthalpies of stable closed-shell polycyclic species calculated at the B3LYP/6-31G(2df,p) level of theory. By combining a molecular graph convolutional neural network with a dropout training strategy, the model we developed can predict density functional theory (DFT) enthalpies for a broad range of polycyclic species and assess the quality of each predicted value. For the species which the current model is uncertain about, the automatic ab initio calculations provide additional training data to improve the performance of the model. For a test set composed of 2858 cyclic and polycyclic hydrocarbons and oxygenates, the enthalpies predicted by the model agree with the reference DFT values with a root-mean-square error of 2.62 kcal/mol. Finally, we found that a model originally trained on hydrocarbons and oxygenates can broaden its prediction coverage to nitrogen-containing species via an active learning process, suggesting that the continuous learning strategy is not only able to improve the model accuracy but is also capable of expanding the predictive capacity of a model to unseen species domains.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗