Machine learning for molecular and materials science

Fast-developing machine learning support complex system research in environmental chemistry

Machine learning will radically accelerate analysis of complex material networks in environmental chemistry.

Towards elucidating structure of ligand-protected nanoclusters

Developing a centralized database for ligand-protected nanoclusters can fuel machine learning and data-science-based approaches towards theoretical structure prediction.

Scalable Topological Data Analysis and Visualization for Evaluating Data-Driven Models in Scientific Applications

With the rapid adoption of machine learning techniques for large-scale applications in science and engineering comes the convergence of two grand challenges in visualization. First, the utilization of black box models (e.g., deep neural networks) calls for advanced techniques in exploring and interpreting model behaviors. Second, the rapid growth in computing has produced enormous datasets that require techniques that can handle millions or more samples. Although some solutions to these interpretability challenges have been proposed, they typically do not scale beyond thousands of samples, nor do they provide the high-level intuition scientists are looking for. Here, we present the first scalable solution to explore and analyze high-dimensional functions often encountered in the scientific data analysis pipeline. By combining a new streaming neighborhood graph construction, the corresponding topology computation, and a novel data aggregation scheme, namely topology aware datacubes, we enable interactive exploration of both the topological and the geometric aspect of high-dimensional data. Following two use cases from high-energy-density (HED) physics and computational biology, we demonstrate how these capabilities have led to crucial new insights in both applications.

Machine Learning Chemical Guidelines for Engineering Electronic Structures in Half-Heusler Thermoelectric Materials

Half-Heusler materials are strong candidates for thermoelectric applications due to their high weighted mobilities and power factors, which is known to be correlated to valley degeneracy in the electronic band structure. However, there are over 50 known semiconducting half-Heusler phases, and it is not clear how the chemical composition affects the electronic structure. While all the n-type electronic structures have their conduction band minimum at either the Γ- or X-point, there is more diversity in the p-type electronic structures, and the valence band maximum can be at either the Γ-, L-, or W-point. Here, we use high throughput computation and machine learning to compare the valence bands of known half-Heusler compounds and discover new chemical guidelines for promoting the highly degenerate W-point to the valence band maximum. We do this by constructing an "orbital phase diagram" to cluster the variety of electronic structures expressed by these phases into groups, based on the atomic orbitals that contribute most to their valence bands. Then, with the aid of machine learning, we develop new chemical rules that predict the location of the valence band maximum in each of the phases. These rules can be used to engineer band structures with band convergence and high valley degeneracy.

Gold-Catalyzed Cross-Coupling Reactions: An Overview of Design Strategies, Mechanistic Studies, and Applications

Transition-metal-catalyzed cross-coupling reactions are central to many organic synthesis methodologies. Traditionally, Pd, Ni, Cu, and Fe catalysts are used to promote these reactions. Recently, many studies have showed that both homogeneous and heterogeneous Au catalysts can be used for activating selective cross-coupling reactions. Here, an overview of the past studies, current trends, and future directions in the field of gold-catalyzed coupling reactions is presented. Design strategies to accomplish selective homocoupling and cross-coupling reactions under both homogeneous and heterogeneous conditions, computational and experimental mechanistic studies, and their applications in diverse fields are critically reviewed. Specific topics covered are: oxidant-assisted and oxidant-free reactions; strain-assisted reactions; dual Au and photoredox catalysis; bimetallic synergistic reactions; mechanisms of reductive elimination processes; enzyme-mimicking Au chemistry; cluster and surface reactions; and plasmonic catalysis. In the relevant sections, theoretical and computational studies of Au^I /Au^III chemistry are discussed and the predictions from the calculations are compared with the experimental observations to derive useful design strategies.

Machine learning based temperature prediction of poly(N-isopropylacrylamide)-capped plasmonic nanoparticle solutions

The temperature-dependent optical properties of gold nanoparticles that are capped with the thermo-sensitive polymer: 'poly(N-isopropylacrylamide)' (PNIPAM), have been studied extensively for several years. Also, their suitability to function as nanoscopic thermometers for bio-sensing applications has been suggested numerous times. In an attempt to establish this, many have studied the temperature-dependent optical resonance characteristics of these particles; however, developing a simple mathematical relationship between the optical measurements and the solution temperature remains an open challenge. In this paper, we attempt to systematically address this problem using machine learning techniques to quickly and accurately predict the solution-temperature, based on spectroscopic data. Our emphasis is on establishing a simple and practically useful solution to this problem. Our dataset comprises spectroscopic absorption data from both nanorods and nanobipyramids capped with PNIPAM, measured at discretely varied and pre-set temperature states. Specific regions of the spectroscopic data are selected as features for prediction using random forest (RF), gradient boosting (GB) and adaptive boosting (AB) regression techniques. Our prediction results indicate that RF and GB techniques can be used successfully to predict solution temperatures instantly to within 1 °C of accuracy.

Deep Learning for Deep Chemistry: Optimizing the Prediction of Chemical Patterns

Computational Chemistry is currently a synergistic assembly between ab initio calculations, simulation, machine learning (ML) and optimization strategies for describing, solving and predicting chemical data and related phenomena. These include accelerated literature searches, analysis and prediction of physical and quantum chemical properties, transition states, chemical structures, chemical reactions, and also new catalysts and drug candidates. The generalization of scalability to larger chemical problems, rather than specialization, is now the main principle for transforming chemical tasks in multiple fronts, for which systematic and cost-effective solutions have benefited from ML approaches, including those based on deep learning (e.g. quantum chemistry, molecular screening, synthetic route design, catalysis, drug discovery). The latter class of ML algorithms is capable of combining raw input into layers of intermediate features, enabling bench-to-bytes designs with the potential to transform several chemical domains. In this review, the most exciting developments concerning the use of ML in a range of different chemical scenarios are described. A range of different chemical problems and respective rationalization, that have hitherto been inaccessible due to the lack of suitable analysis tools, is thus detailed, evidencing the breadth of potential applications of these emerging multidimensional approaches. Focus is given to the models, algorithms and methods proposed to facilitate research on compound design and synthesis, materials design, prediction of binding, molecular activity, and soft matter behavior. The information produced by pairing Chemistry and ML, through data-driven analyses, neural network predictions and monitoring of chemical systems, allows (i) prompting the ability to understand the complexity of chemical data, (ii) streamlining and designing experiments, (ii) discovering new molecular targets and materials, and also (iv) planning or rethinking forthcoming chemical challenges. In fact, optimization engulfs all these tasks directly.

Finding Reactive Configurations: A Machine Learning Approach for Estimating Energy Barriers Applied to Sirtuin 5

Sirtuin 5 is a class III histone deacetylase that, unlike its classification, mainly catalyzes desuccinylation and demanoylation reactions. It is an interesting drug target that we use here to test new ideas for calculating reaction pathways of large molecular systems such as enzymes. A major issue with most schemes (e.g., adiabatic mapping) is that the resulting activation barrier height heavily depends on the chosen educt conformation. This makes the selection of the initial structure decisive for the success of the characterization. Here, we apply machine learning to a large number of molecular dynamics frames and potential energy barriers obtained by quantum mechanics/molecular mechanics calculations in order to identify (1) suitable start-conformations for reaction path calculations and (2) structural features relevant for the first step of the desuccinylation reaction catalyzed by Sirtuin 5. The latter generally aids the understanding of reaction mechanisms and important interactions in active centers. Using our novel approach, we found eleven key features that govern the reactivity. We were able to estimate reaction barriers with a mean absolute error of 3.6 kcal/mol and identified reactive configurations.

The inverse design of structural color using machine learning

Efficiently identifying optical structures with desired functionalities, referred to as inverse design, can dramatically accelerate the invention of new photonic devices, and this is especially useful in the design of large scale integrated photonic chips. Structural color with high-resolution, high-saturation, and low-loss holds great promise in image display, data storage and information security. However, the inverse design of structural color remains an open challenge, and this impedes practical application. Here, we propose an inverse design strategy for structural color using machine learning (ML) technologies. The supervised learning (SL) models are trained with the geometries and colors of dielectric arrays to capture accurate geometry-color relationships, and these are then applied to a reinforcement learning (RL) algorithm in order to find the optical structural geometries for the desired color. Our work succeeds in finding simple and accurate models to describe geometry-color relationships, which significantly improves the efficiency of the design. This strategy provides a systematic method to directly encode generic functionality into a set of structures and geometries, paving the way for the inverse design of functional photonic devices.

Unsupervised discovery of solid-state lithium ion conductors

Although machine learning has gained great interest in the discovery of functional materials, the advancement of reliable models is impeded by the scarcity of available materials property data. Here we propose and demonstrate a distinctive approach for materials discovery using unsupervised learning, which does not require labeled data and thus alleviates the data scarcity challenge. Using solid-state Li-ion conductors as a model problem, unsupervised materials discovery utilizes a limited quantity of conductivity data to prioritize a candidate list from a wide range of Li-containing materials for further accurate screening. Our unsupervised learning scheme discovers 16 new fast Li-conductors with conductivities of 10⁻⁴–10⁻¹ S cm⁻¹ predicted in ab initio molecular dynamics simulations. These compounds have structures and chemistries distinct to known systems, demonstrating the capability of unsupervised learning for discovering materials over a wide materials space with limited property data.

Predictions of new solid-state Li-ion conductors are challenging due to the diverse chemistries and compositions involved. Here the authors combine unsupervised learning techniques and molecular dynamics simulations to discover new compounds with high Li-ion conductivity.

Virtual Materials Intelligence for Design and Discovery of Advanced Electrocatalysts

Similar to advancements gained from big data in genomics, security, internet of things, and e-commerce, the materials workflow could be made more efficient and prolific through advances in streamlining data sources, autonomous materials synthesis, rapid characterization, big data analytics, and self-learning algorithms. In electrochemical materials science, data sets are large, unstructured/heterogeneous, and difficult to process and analyze from a single data channel or platform. Computer-aided materials design together with advances in data mining, machine learning, and predictive analytics are expected to provide inexpensive and accelerated pathways towards tailor-made functionally optimized energy materials. Fundamental research in the field of electrochemical energy materials focuses primarily on complex interfacial phenomena and kinetic electrocatalytic processes. This perspective article critically assesses AI-driven modeling and computational approaches that are currently applied to those objects. An application-driven materials intelligence platform is introduced, and its functionalities are scrutinized considering the development of electrocatalyst materials for CO₂ conversion as a use case.

Prediction of Absorption Spectrum Shifts in Dyes Adsorbed on Titania

Dye adsorption on metal-oxide films often results in small to substantial absorption shifts relative to the solution phase, with undesirable consequences for the performance of dye-sensitized solar cells and optical sensors. While density functional theory is frequently used to model such behaviour, it is too time-consuming for rapid assessment. In this paper, we explore the use of supervised machine learning to predict whether dye adsorption on titania is likely to induce a change in its absorption characteristics. The physicochemical features of each dye were encoded as a numeric vector whose elements are the counts of molecular fragments and topological indices. Various classification models were subsequently trained to predict the type of absorption shift i.e. blue, red or unchanged (|Δλ| ≤ 10 nm). The models were able to predict the nature of the shift with a good likelihood (~80%) of success when applied to unseen data.

Machine learning for target discovery in drug development

The discovery of macromolecular targets for bioactive agents is currently a bottleneck for the informed design of chemical probes and drug leads. Typically, activity profiling against genetically manipulated cell lines or chemical proteomics is pursued to shed light on their biology and deconvolute drug-target networks. By taking advantage of the ever-growing wealth of publicly available bioactivity data, learning algorithms now provide an attractive means to generate statistically motivated research hypotheses and thereby prioritize biochemical screens. Here, we highlight recent successes in machine intelligence for target identification and discuss challenges and opportunities for drug discovery.

Representing Multiword Chemical Terms through Phrase-Level Preprocessing and Word Embedding

In recent years, data-driven methods and artificial intelligence have been widely used in chemoinformatic and material informatics domains, for which the success is critically determined by the availability of training data with good quality and large quantity. A potential approach to break this bottleneck is by leveraging the chemical literature such as papers and patents as alternative data resources to high throughput experiments and simulation. Compared to other domains where natural language processing techniques have established successes, the chemical literature contains a large portion of phrases of multiple words that create additional challenges for accurate identification and representation. Here, we introduce a chemistry domain suitable approach to identify multiword chemical terms and train word representations at the phrase level. Through a series of special-designed experiments, we demonstrate that our multiword identifying and representing method effectively and accurately identifies multiword chemical terms from 119, 166 chemical patents and is more robust and precise to preserve the semantic meaning of chemical phrases compared to the conventional approach, which represents constituent single words first and combine them afterward. Because the accurate representation of chemical terms is the first and essential step to provide learning features for downstream natural language processing tasks, our results pave the road to utilize the large volume of chemical literature in future data-driven studies.

Alloyed Thermoelectric PbTe-SnTe Films Formed via Aerosol Deposition

The discovery of new thermoelectric materials has the potential to benefit from advances in high-throughput methodologies. Traditional synthesis and characterization routes for thermoelectrics are time-consuming serial processes. In contrast, high-throughput materials discovery is commonly done by thin film growth, which may produce microstructures that are metastable or compositionally graded and, therefore, are challenging to characterize. As a middle ground between bulk synthesis and thin film deposition, we find that the aerosol deposition process can rapidly produce samples that exhibit electronic property trends consistent with those produced by traditional bulk means. We demonstrate rapid growth of discrete thermoelectric thick films of varying chemical compositions (Pb_1-xSn_xTe) from PbTe and SnTe polydisperse micrometer sized powder feedstocks. The high deposition rate (near 1 μm min^-1) and resultant microstructures are advantageous as the diffusion length scales promote rapid thermal treatment and equilibrium phase formation. Room-temperature high-throughput measurements of the Seebeck coefficient and resistivity are compared to traditionally produced bulk materials. The Seebeck coefficient of the films follows the trends of traditional samples, but the resistivity is found to be more sensitive to microstructural effects. Ultimately, we demonstrate a framework for exploratory materials science using aerosol deposition and high-throughput characterization instrumentation.

Data-Driven Materials Science: Status, Challenges, and Perspectives

Data-driven science is heralded as a new paradigm in materials science. In this field, data is the new resource, and knowledge is extracted from materials datasets that are too big or complex for traditional human reasoning-typically with the intent to discover new or improved materials or materials phenomena. Multiple factors, including the open science movement, national funding, and progress in information technology, have fueled its development. Such related tools as materials databases, machine learning, and high-throughput methods are now established as parts of the materials research toolset. However, there are a variety of challenges that impede progress in data-driven materials science: data veracity, integration of experimental and computational data, data longevity, standardization, and the gap between industrial interests and academic efforts. In this perspective article, the historical development and current state of data-driven materials science, building from the early evolution of open science to the rapid expansion of materials data infrastructures are discussed. Key successes and challenges so far are also reviewed, providing a perspective on the future development of the field.

Accelerated Search for BaTiO3-Based Ceramics with Large Energy Storage at Low Fields Using Machine Learning and Experimental Design

The problem that is considered is that of maximizing the energy storage density of Pb-free BaTiO3-based dielectrics at low electric fields. It is demonstrated that how varying the size of the combinatorial search space influences the efficiency of material discovery by comparing the performance of two machine learning based approaches where different levels of physical insights are involved. It is started with physics intuition to provide guiding principles to find better performers lying in the crossover region in the composition-temperature phase diagram between the ferroelectric phase and relaxor ferroelectric phase. Such an approach is limiting for multidopant solid solutions and motivates the use of two data-driven machine learning and design strategies with a feedback loop to experiments. Strategy I considers learning and property prediction on all the compounds, and strategy II learns to preselect compounds in the crossover region on which prediction is carried out. By performing only two active learning loops via strategy II, the compound (Ba0.86Ca0.14)(Ti0.79Zr0.11Hf0.10)O3 is synthesized with the largest energy storage density ≈73 mJ cm-3 at a field of 20 kV cm-1, and an insight into the relative performance of the strategies using varying levels of knowledge is provided.

Machine Learning Interatomic Potentials as Emerging Tools for Materials Science

Atomic-scale modeling and understanding of materials have made remarkable progress, but they are still fundamentally limited by the large computational cost of explicit electronic-structure methods such as density-functional theory. This Progress Report shows how machine learning (ML) is currently enabling a new degree of realism in materials modeling: by "learning" electronic-structure data, ML-based interatomic potentials give access to atomistic simulations that reach similar accuracy levels but are orders of magnitude faster. A brief introduction to the new tools is given, and then, applications to some select problems in materials science are highlighted: phase-change materials for memory devices; nanoparticle catalysts; and carbon-based electrodes for chemical sensing, supercapacitors, and batteries. It is hoped that the present work will inspire the development and wider use of ML-based interatomic potentials in diverse areas of materials research.

Evolutionary approaches in protein engineering towards biomaterial construction

The tailoring of proteins for specific applications by evolutionary methods is a highly active area of research. Rational design and directed evolution are the two main strategies to reengineer proteins or create chimeric structures. Rational engineering is often limited by insufficient knowledge about proteins' structure-function relationships; directed evolution overcomes this restriction but poses challenges in the screening of candidates. A combination of these protein engineering approaches will allow us to create protein variants with a wide range of desired properties. Herein, we focus on the application of these approaches towards the generation of protein biomaterials that are known for biodegradability, biocompatibility and biofunctionality, from combinations of natural, synthetic, or engineered proteins and protein domains. Potential applications depend on the enhancement of biofunctional, mechanical, or other desired properties. Examples include scaffolds for tissue engineering, thermostable enzymes for industrial biocatalysis, and other therapeutic applications.

QM-sym, a symmetrized quantum chemistry database of 135 kilo molecules

Applying deep learning methods in materials science research is an important way of solving the time-consuming problems of typical ab initio quantum chemistry methodology, but due to the size of large molecules, large and uncharted fields still exist. Implementing symmetry information can significantly reduce the calculation complexity of structures, as they can be simplified to the minimum symmetric units. Because there are few quantum chemistry databases that include symmetry information, we constructed a new one, named QM-sym, by designing an algorithm to generate 135k organic molecules with the Cnh symmetry composite. Those generated molecules were optimized to a stable state using Gaussian 09. The geometric, electronic, energetic, and thermodynamic properties of the molecules were calculated, including their orbital degeneracy states and orbital symmetry around the HOMO-LUMO. The basic symmetric units were also included. This database p rovides consistent and comprehensive quantum chemical properties for structures with Cnh symmetries. QM-sym can be used as a benchmark for machine learning models in quantum chemistry or as a dataset for training new symmetry-based models.

Machine Learning the Physical Nonlocal Exchange-Correlation Functional of Density-Functional Theory

We train a neural network as the universal exchange-correlation functional of density-functional theory that simultaneously reproduces both the exact exchange-correlation energy and the potential. This functional is extremely nonlocal but retains the computational scaling of traditional local or semilocal approximations. It therefore holds the promise of solving some of the delocalization problems that plague density-functional theory, while maintaining the computational efficiency that characterizes the Kohn-Sham equations. Furthermore, by using automatic differentiation, a capability present in modern machine-learning frameworks, we impose the exact mathematical relation between the exchange-correlation energy and the potential, leading to a fully consistent method. We demonstrate the feasibility of our approach by looking at one-dimensional systems with two strongly correlated electrons, where density-functional methods are known to fail, and investigate the behavior and performance of our functional by varying the degree of nonlocality.

Text-mined dataset of inorganic materials synthesis recipes

Materials discovery has become significantly facilitated and accelerated by high-throughput ab-initio computations. This ability to rapidly design interesting novel compounds has displaced the materials innovation bottleneck to the development of synthesis routes for the desired material. As there is no a fundamental theory for materials synthesis, one might attempt a data-driven approach for predicting inorganic materials synthesis, but this is impeded by the lack of a comprehensive database containing synthesis processes. To overcome this limitation, we have generated a dataset of “codified recipes” for solid-state synthesis automatically extracted from scientific publications. The dataset consists of 19,488 synthesis entries retrieved from 53,538 solid-state synthesis paragraphs by using text mining and natural language processing approaches. Every entry contains information about target material, starting compounds, operations used and their conditions, as well as the balanced chemical equation of the synthesis reaction. The dataset is publicly available and can be used for data mining of various aspects of inorganic materials synthesis.

Measurement(s)	solid-state synthesis data
Technology Type(s)	natural language processing

Machine-accessible metadata file describing the reported data: 10.6084/m9.figshare.9906608

Machine Learning Applied to Zeolite Synthesis: The Missing Link for Realizing High-Throughput Discovery

Zeolites are microporous crystalline materials with well-defined cavities and pores, which can be prepared under different pore topologies and chemical compositions. Their preparation is typically defined by multiple interconnected variables (e.g., reagent sources, molar ratios, aging treatments, reaction time and temperature, among others), but unfortunately their distinctive influence, particularly on the nucleation and crystallization processes, is still far from being understood. Thus, the discovery and/or optimization of specific zeolites is closely related to the exploration of the parametric space through trial-and-error methods, generally by studying the influence of each parameter individually. In the past decade, machine learning (ML) methods have rapidly evolved to address complex problems involving highly nonlinear or massively combinatorial processes that conventional approaches cannot solve. Considering the vast and interconnected multiparametric space in zeolite synthesis, coupled with our poor understanding of the mechanisms involved in their nucleation and crystallization, the use of ML is especially timely for improving zeolite synthesis. Indeed, the complex space of zeolite synthesis requires drawing inferences from incomplete and imperfect information, for which ML methods are very well-suited to replace the intuition-based approaches traditionally used to guide experimentation. In this Account, we contend that both existing and new ML approaches can provide the "missing link" needed to complete the traditional zeolite synthesis workflow used in our quest to rationalize zeolite synthesis. Within this context, we have made important efforts on developing ML tools in different critical areas, such as (1) data-mining tools to process the large amount of data generated using high-throughput platforms; (2) novel complex algorithms to predict the formation of energetically stable hypothetical zeolites and guide the synthesis of new zeolite structures; (3) new "ab initio" organic structure directing agent predictions to direct the synthesis of hypothetical or known zeolites; (4) an automated tool for nonsupervised data extraction and classification from published research articles. ML has already revolutionized many areas in materials science by enhancing our ability to map intricate behavior to process variables, especially in the absence of well-understood mechanisms. Undoubtedly, ML is a burgeoning field with many future opportunities for further breakthroughs to advance the design of molecular sieves. For this reason, this Account includes an outlook of future research directions based on current challenges and opportunities. We envision this Account will become a hallmark reference for both well-established and new researchers in the field of zeolite synthesis.

Statistical learning goes beyond the d-band model providing the thermochemistry of adsorbates on transition metals

The rational design of heterogeneous catalysts relies on the efficient survey of mechanisms by density functional theory (DFT). However, massive reaction networks cannot be sampled effectively as they grow exponentially with the size of reactants. Here we present a statistical principal component analysis and regression applied to the DFT thermochemical data of 71 C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${}_{1}$$\end{document}1–C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${}_{2}$$\end{document}2 species on 12 close-packed metal surfaces. Adsorption is controlled by covalent (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d$$\end{document}d-band center) and ionic terms (reduction potential), modulated by conjugation and conformational contributions. All formation energies can be reproduced from only three key intermediates (predictors) calculated with DFT. The results agree with accurate experimental measurements having error bars comparable to those of DFT. The procedure can be extended to single-atom and near-surface alloys reducing the number of explicit DFT calculation needed by a factor of 20, thus paving the way for a rapid and accurate survey of whole reaction networks on multimetallic surfaces.

Assessing catalytic mechanisms using DFT calculations greatly aids catalyst design, but is impractical for large molecules. Here the authors develop a statistical learning-based thermochemical model for estimating adsorption of organics onto metals, retaining DFT accuracy while reducing the number of calculations by a factor of 20.

Revisiting the Potential Energy Surface of the Stacked Cytosine Dimer: FNO-CCSD(T) Interaction Energies, SAPT Decompositions, and Benchmarking

Nucleobase stacking interactions are crucial for the stability of nucleic acids. This study investigates base stacking energies of the cytosine homodimer in different configurations, including intermolecular separation plots, detailed twist dependence, and displaced structures. Highly accurate ab initio quantum chemical single point energies using an energy function based on MP2 complete basis set extrapolation ([6 → 7]ZaPa-NR) and a CCSD(T)/cc-pVTZ-F12 high-level correction are presented as new reference data, providing the most accurate stacking energies of nucleobase dimers currently available. Accurate SAPT2+(3)δMP2 energy decomposition is used to obtain detailed insights into the nature of base stacking interactions at varying vertical distances and twist values. The ab initio symmetry adapted perturbation theory (SAPT) energy decomposition suggests that the base stacking originates from an intricate interplay between dispersion attraction, short-range exchange-repulsion, and Coulomb interaction. The interpretation of the SAPT data is a complex issue as key energy terms vary substantially in the region of optimal (low energy) base stacking geometries. Thus, attempts to highlight one leading stabilizing SAPT base stacking term may be misleading and the outcome strongly depends on the used geometries within the range of geometries sampled in nucleic acids upon thermal fluctuations. Modern dispersion-corrected density functional theory (among them DSD-BLYP-D3, ωB97M-V, and ωB97M-D3BJ) is benchmarked and often reaches up to spectroscopic accuracy (below 1 kJ/mol). The classical AMBER force field is benchmarked with multiple different sets of point-charges (e.g. HF, DFT, and MP2-based) and is found to produce reasonable agreement with the benchmark data.

Prediction of learning curves of 2 dental CAD software programs, part 2: Differences in learning effects by type of dental personnel

STATEMENT OF PROBLEM

Dental computer-aided design (CAD) software programs are essential elements of the digital workflow. Therefore, it is necessary to study the learning effect of dental CAD software programs for efficient use.

PURPOSE

The purpose of this in vitro study was to predict the learning curve of dental CAD software programs according to dental personnel by using the Wright model and to investigate the tendency of dental personnel to reduce working time according to repeated learning.

MATERIAL AND METHODS

A total of 36 participants were recruited, including an equal number of dentists, dental technicians, and dental students (12 each). A custom abutment design was evaluated by using exocad CAD and Deltanine CAD software programs. The design was carried out in the following order: 4 steps repeated 3 times each. This study applied the formula of the Wright model to predict 500 repetitive times. In the statistical analysis, 3-repetition and 500-repetition times were analyzed with the Kruskal-Wallis H test and Friedman test (α=.05), and a post hoc comparison was performed by using the Mann-Whitney U-test and Bonferroni correction method (α=.017).

RESULTS

Three repetitions resulted in shorter working time in the dental technician group. The 3-repetition time decreased statistically for all dental personnel (P<.001). The time for 500 repetitions showed a statistically significant difference according to the type of dental personnel (P=.036), but no significant difference was found after the fourth iteration (fifth iteration: P=.076). Furthermore, the estimated time of 500 iterations decreased statistically significantly from the first to the 500th iteration (P<.001).

CONCLUSIONS

All dental personnel showed learning effects of dental CAD software programs. Although the dental technician group initially showed less working time, after initial learning, the same learning effect appeared, regardless of the type of dental personnel.

Prediction and optimization of epoxy adhesive strength from a small dataset through active learning

Machine learning is emerging as a powerful tool for the discovery of novel high-performance functional materials. However, experimental datasets in the polymer-science field are typically limited and they are expensive to build. Their size (< 100 samples) limits the development of chemical intuition from experimentalists, as it constrains the use of machine-learning algorithms for extracting relevant information. We tackle this issue to predict and optimize adhesive materials by combining laboratory experimental design, an active learning pipeline and Bayesian optimization. We start from an initial dataset of 32 adhesive samples that were prepared from various molecular-weight bisphenol A-based epoxy resins and polyetheramine curing agents, mixing ratios and curing temperatures, and our data-driven method allows us to propose an optimal preparation of an adhesive material with a very high adhesive joint strength measured at 35.8 ± 1.1 MPa after three active learning cycles (five proposed preparations per cycle). A Gradient boosting machine learning model was used for the successive prediction of the adhesive joint strength in the active learning pipeline, and the model achieved a respectable accuracy with a coefficient of determination, root mean square error and mean absolute error of 0.85, 4.0 MPa and 3.0 MPa, respectively. This study demonstrates the important impact of active learning to accelerate the design and development of tailored highly functional materials from very small datasets.

Computation-Guided Design of Ni-Mn-Sn Ferromagnetic Shape Memory Alloy with Giant Magnetocaloric Effect and Excellent Mechanical Properties and High Working Temperature via Multielement Doping

Ni-Mn-Sn ferromagnetic shape memory alloys (FSMAs) have promise for application in efficient solid-state refrigeration. However, the simultaneous achievement of giant magnetocaloric effect (MCE) and excellent mechanical properties and high working temperature in these materials is always the challenge. Computation-guided materials design techniques provide an efficient way to design and identify new magnetocaloric materials. Herein, a new strategy of multidoping is presented. First, we conduct a detailed and comprehensive first-principles study and predict that Ni-Mn-Sn FSMAs with co-doping 6.25 atom % Cu and 6.25-12.5 atom % Co can realize the multiobjective optimization of magnetocaloric material. Then, it is confirmed by experiment and we report on Ni₄₀Co₈Mn₃₇Sn₉Cu₆ FSMA exhibiting a large magnetic entropy change (34.8 J/(kg K)) of a large value in the prevalent MCE materials at high temperature (∼344 K) and whose compression stress and strain (∼1072.0 MPa and ∼11.9%) are both the largest among Ni-Mn-based MCE materials. Notably, the effect of Co and Cu doping is not simply stacked because they play opposite roles in Curie temperature (T_C) and martensitic transformation temperature (T_M). So, achieving the balance of their effect to combine their merits in a very narrow window is the key step. This approach of multielement doping holds promise to be extended to other magnetocaloric materials to enhance their multiple properties simultaneously.

Machine learning enables long time scale molecular photodynamics simulations

Photo-induced processes are fundamental in nature but accurate simulations of their dynamics are seriously limited by the cost of the underlying quantum chemical calculations, hampering their application for long time scales. Here we introduce a method based on machine learning to overcome this bottleneck and enable accurate photodynamics on nanosecond time scales, which are otherwise out of reach with contemporary approaches. Instead of expensive quantum chemistry during molecular dynamics simulations, we use deep neural networks to learn the relationship between a molecular geometry and its high-dimensional electronic properties. As an example, the time evolution of the methylenimmonium cation for one nanosecond is used to demonstrate that machine learning algorithms can outperform standard excited-state molecular dynamics approaches in their computational efficiency while delivering the same accuracy.

Phillips-Inspired Machine Learning for Band Gap and Exciton Binding Energy Prediction

In this work, inspired by Phillips's ionicity theory in solid-state physics, we directly sort out the critical factors of the band gap's feature correlations in the machine learning architected with the Lasso algorithm. Even based on a small 2D materials data set, we can fundamentally approach an accurate and rational model about the band gap and exciton binding energy with robust transferability to other databases. Our machine learning outputs can reveal the exact physics pictures behind the predicted quantity as well as the "secondary understanding" of the correlation between the approximated physics models in exciton. This work stresses the significant value of physics endorsement on the machine learning (ML) algorithm and provides a symbolic regression solution for the "few-shot" training scheme for ML technology in materials science. Moreover, physics-inspired secondary understanding could be an essential supplement for ML in scientific research fields.

Regulating, Measuring, and Modeling the Viscoelasticity of Bacterial Biofilms

Biofilms occur in a broad range of environments under heterogeneous physicochemical conditions, such as in bioremediation plants, on surfaces of biomedical implants, and in the lungs of cystic fibrosis patients. In these scenarios, biofilms are subjected to shear forces, but the mechanical integrity of these aggregates often prevents their disruption or dispersal. Biofilms' physical robustness is the result of the multiple biopolymers secreted by constituent microbial cells which are also responsible for numerous biological functions. A better understanding of the role of these biopolymers and their response to dynamic forces is therefore crucial for understanding the interplay between biofilm structure and function. In this paper, we review experimental techniques in rheology, which help quantify the viscoelasticity of biofilms, and modeling approaches from soft matter physics that can assist our understanding of the rheological properties. We describe how these methods could be combined with synthetic biology approaches to control and investigate the effects of secreted polymers on the physical properties of biofilms. We argue that without an integrated approach of the three disciplines, the links between genetics, composition, and interaction of matrix biopolymers and the viscoelastic properties of biofilms will be much harder to uncover.

A quantitative uncertainty metric controls error in neural network-driven chemical discovery

Machine learning (ML) models, such as artificial neural networks, have emerged as a complement to high-throughput screening, enabling characterization of new compounds in seconds instead of hours. The promise of ML models to enable large-scale chemical space exploration can only be realized if it is straightforward to identify when molecules and materials are outside the model's domain of applicability. Established uncertainty metrics for neural network models are either costly to obtain (e.g., ensemble models) or rely on feature engineering (e.g., feature space distances), and each has limitations in estimating prediction errors for chemical space exploration. We introduce the distance to available data in the latent space of a neural network ML model as a low-cost, quantitative uncertainty metric that works for both inorganic and organic chemistry. The calibrated performance of this approach exceeds widely used uncertainty metrics and is readily applied to models of increasing complexity at no additional cost. Tightening latent distance cutoffs systematically drives down predicted model errors below training errors, thus enabling predictive error control in chemical discovery or identification of useful data points for active learning.

Efficient multi-objective molecular optimization in a continuous latent space

One of the main challenges in small molecule drug discovery is finding novel chemical compounds with desirable properties. In this work, we propose a novel method that combines in silico prediction of molecular properties such as biological activity or pharmacokinetics with an in silico optimization algorithm, namely Particle Swarm Optimization. Our method takes a starting compound as input and proposes new molecules with more desirable (predicted) properties. It navigates a machine-learned continuous representation of a drug-like chemical space guided by a defined objective function. The objective function combines multiple in silico prediction models, defined desirability ranges and substructure constraints. We demonstrate that our proposed method is able to consistently find more desirable molecules for the studied tasks in relatively short time. We hope that our method can support medicinal chemists in accelerating and improving the lead optimization process.

Automatically Identifying Electrode Reaction Mechanisms Using Deep Neural Networks

At present, electrochemical mechanisms are most commonly identified subjectively based on the experience of the researcher. This subjectivity is reflected in bias to particular mechanisms as well as lack of quantifiable confidence in the chosen mechanism compared to potential alternative mechanisms. In this paper we demonstrate that a deep neural network trained to recognize dc cyclic voltammograms for three commonly encountered mechanisms provides correct classifications within 5 ms without the problem of subjectivity. To mimic experimental data, the impact of noise, uncompensated resistance, and dependence on scan rate, factors that are relevant to practical studies, has also been investigated. Outcomes with two experimental data sets are also presented.

Predicting the onset temperature (Tg) of GexSe1-x glass transition: a feature selection based two-stage support vector regression method

Despite the usage of both experimental and topological methods, realizing a rapid and accurate measurement of the onset temperature (T_g) of Ge_xSe_1-x glass transition remains an open challenge. In this paper, a predictive model for the T_g in Ge_xSe_1-x glass system is presented by a machine learning method named feature selection based two-stage support vector regression (FSTS-SVR). Firstly, Pearson correlation coefficient (PCC) is used to select features highly correlated with T_g from the candidate features of Ge_xSe_1-x glass system. Secondly, in order to simulate the two-stage characteristic of T_g which is caused by structural variation with a turning point at x = 0.33 via the structural analysis, SVR is utilized to build predictive models for two stages separately and then the two achieved models are synthesized using a minimum error based model for T_g prediction. Compared with the topological and other methods based on SVR, the FSTS-SVR gives the highest predictive accuracy with the root mean square error (RMSE) and mean absolute percentage error (MAPE) of 10.64 K and 2.38%, respectively. This method is also expected to be more efficient for the prediction of T_g of other glass systems with the multi-stage characteristic.

Enhanced Deep-Learning Prediction of Molecular Properties via Augmentation of Bond Topology

Deep learning has made great strides in tackling chemical problems, but still lacks full-fledged representations for three-dimensional (3D) molecular structures for its inner working. For example, the molecular graph, commonly used in chemistry and recently adapted to the graph convolutional network (GCN), is inherently a 2D representation of 3D molecules. Herein we propose an advanced version of the GCN, called 3DGCN, which receives 3D molecular information from a molecular graph augmented by information on bond direction. While outperforming state-of-the-art deep-learning models in the prediction of chemical and biological properties, 3DGCN has the ability to both generalize and distinguish molecular rotations in 3D, beyond 2D, which has great impact on drug discovery and development, not to mention the design of chemical reactions.