Reaction Templates: Bridging Synthesis Knowledge and Artificial Intelligence.

ConspectusThe field of chemical research boasts a long history of developing software to automate synthesis planning and reaction prediction. Early software relied heavily on expert systems, requiring significant effort to encode vast amounts of synthesis knowledge into a computer-readable format. However, recent advancements in deep learning have shifted the focus toward AI models, offering improved prediction capabilities. Despite these advancements, current AI models often lack the integration of known synthesis rules and intuitions, creating a gap that hinders interpretability and future development of the models. To bridge them, our research group has been actively working on incorporating reaction templates into deep learning models, achieving promising results across various applications.In this Account, we present our latest works to incorporate the known synthesis knowledge into the deep learning models through the utilization of reaction templates. We begin by highlighting the limitations of early computer programs heavily reliant on hand-coded rules. These programs, while providing a foundation for the field, presented limitations in scalability and adaptability. We then introduce SMARTS (SMILES arbitrary target specification), a popular Python-readable format for representing chemical reactions. This format of reaction encoding facilitates the quick integration of synthesis knowledge into AI models built using the Python language. With the SMARTS-based reaction templates, we introduce our recent efforts of developing an AI model for reaction-based molecule optimization. Subsequently, we discuss the recent efforts to automate the extraction of reaction templates from vast chemical reaction databases. This approach eliminates the previously required manual effort of encoding knowledge, a process that could be time-consuming and prone to error when dealing with large data sets. By customizing the automated extraction algorithm, we have developed powerful AI models for specific tasks such as retrosynthesis (LocalRetro), reaction outcome prediction (LocalTransform), and atom-to-atom mapping (LocalMapper). These models, aligned with the intuition of chemists, demonstrate the effectiveness of incorporating reaction templates into deep learning frameworks.Looking toward the future, we believe that utilizing reaction templates to connect known chemical knowledge and AI models holds immense potential for various applications. Not only can this approach significantly benefit future AI models focused on challenging tasks like reaction mechanism labeling and prediction, but we anticipate it can also extend its reach to the realm of inorganic synthesis. By integrating synthesis knowledge, we can not only achieve improved performance but also enhance the interpretability of AI models, paving the way for further advancements in AI-powered chemical synthesis.

Variational quantum eigensolver for closed-shell molecules with non-bosonic corrections.

The realization of quantum advantage with noisy-intermediate-scale quantum (NISQ) machines has become one of the major challenges in computational sciences. Maintaining coherence of a physical system with more than ten qubits is a critical challenge that motivates research on compact system representations to reduce algorithm complexity. Toward this end, the variational quantum eigensolver (VQE) used to perform quantum simulations is considered to be one of the most promising algorithms for quantum chemistry in the NISQ era. We investigate reduced mapping of one spatial orbital to a single qubit to analyze the ground state energy in a way that the Pauli operators of qubits are mapped to the creation/annihilation of singlet pairs of electrons. To include the effect of non-bosonic (or non-paired) excitations, we introduce a simple correction scheme in the electron correlation model approximated by the geometrical mean of the bosonic (or paired) terms. Employing it in a VQE algorithm, we assess ground state energies of H2O, N2, and Li2O in good agreement with full configuration interaction (FCI) models respectively, using only 6, 8, and 12 qubits with quantum gate depths proportional to the squares of the qubit counts. With the adopted seniority-zero approximation that uses only one half of the qubit counts of a conventional VQE algorithm, we find that our non-bosonic correction method reaches reliable quantum chemistry simulations at least for the tested systems.

Exploring the Diverse Landscape of Biaryl-Containing Peptides Generated by Cytochrome P450 Macrocyclases.

Cytochrome P450 enzymes (P450s) catalyze diverse oxidative cross-coupling reactions between aromatic substrates in the natural product biosynthesis. Specifically, P450s install distinct biaryl macrocyclic linkages in three families of ribosomally synthesized and post-translationally modified peptides (RiPPs). However, the chemical diversity of biaryl-containing macrocyclic RiPPs remains largely unexplored. Here, we demonstrate that P450s have the capability to generate diverse biaryl linkages on RiPPs, collectively named "cyptides". Homology-based genome mining for P450 macrocyclases revealed 19 novel groups of homologous biosynthetic gene clusters (BGCs) with distinct aromatic residue patterns in the precursor peptides. Using the P450-modified precursor peptides heterologously coexpressed with corresponding P450s in Escherichia coli, we determined the NMR structures of three novel biaryl-containing peptides─the enzymatic products, roseovertin (1), rubrin (2), and lapparbin (3)─and confirmed the formation of three unprecedented or rare biaryl linkages: Trp C-7'-to-His N-τ in 1, Trp C-7'-to-Tyr C-6 in 2, and Tyr C-6-to-Trp N-1' in 3. Biochemical characterization indicated that certain P450s in these pathways have a relaxed substrate specificity. Overall, our studies suggest that P450 macrocyclases have evolved to create diverse biaryl linkages in RiPPs, promoting the exploration of a broader chemical space for biaryl-containing peptides encoded in bacterial genomes.

Correlated Local Fluctuations in the Hydrogen Bond Network of Liquid Water.

The hypothesis that liquid water can separate into two phases in the supercooled state has been supported by recent experimental and theoretical studies. However, whether such structural inhomogeneity extends to ambient conditions is under intense debate. Due to the dynamic nature of the hydrogen bond network of liquid water, exploring its structure requires detailed insight into the collective motion of neighboring water molecules, a missing link that has not been examined so far. Here, highly sensitive quantum mechanical calculations detect that the time evolution of nearby hydrogen bonds is strongly correlated, revealing a direct mechanism for the appearance of short-range structural fluctuations in the hydrogen bond network of liquid water for the first time. This correlated dynamics is found to be closely connected to the static structural picture. The distortions from the tetrahedral structure do not occur independently but are correlated due to the preference of nearby donors and acceptors to be in similar environments. The existence of such cooperative fluctuations is further supported by the temperature dependence of the local structural evolution and explained by conventional analysis of localized orbitals. It was found that such correlated structural fluctuations are only observed on a short length scale in simulations at ambient conditions. The correlations of the nearby hydrogen bond pairs of liquid water unveiled here are expected to offer a new insight into connecting the dynamics of individual water molecules and the local structure of the hydrogen bond network.

Single-Atom Molybdenum-N3 Sites for Selective Hydrogenation of CO2 to CO.

The design of efficient non-noble metal catalysts for CO2 hydrogenation to fuels and chemicals is desired yet remains a challenge. Herein, we report that single Mo atoms with a MoN3 (pyrrolic) moiety enable remarkable CO2 adsorption and hydrogenation to CO, as predicted by density functional theory studies and evidenced by a high and stable conversion of CO2 reaching about 30.4 % with a CO selectivity of almost 100 % at 500 °C and very low H2 partial pressure. Atomically dispersed MoN3 is calculated to facilitate CO2 activation and reduces CO2 to CO* via the direct dissociation path. Furthermore, the highest transition state energy in CO formation is 0.82 eV, which is substantially lower than that of CH4 formation (2.16 eV) and accounts for the dominant yield of CO. The enhanced catalytic performances of Mo/NC originate from facile CO desorption with the help of dispersed Mo on nitrogen-doped carbon (Mo/NC), and in the absence of Mo nanoparticles. The resulting catalyst preserves good stability without degradation of CO2 conversion rate even after 68 hours of continuous reaction. This finding provides a promising route for the construction of highly active, selective, and robust single-atom non-precious metal catalysts for reverse water-gas shift reaction.

Autobifunctional Mechanism of Jagged Pt Nanowires for Hydrogen Evolution Kinetics via End-to-End Simulation.

The extraordinary mass activity of jagged Pt nanowires can substantially improve the economics of the hydrogen evolution reaction (HER). However, it is a great challenge to fully unveil the HER kinetics driven by the jagged Pt nanowires with their multiscale morphology. Herein we present an end-to-end framework that combines experiment, machine learning, and multiscale advances of the past decade to elucidate the HER kinetics catalyzed by jagged Pt nanowires under alkaline conditions. The bifunctional catalysis conventionally refers to the synergistic increase in activity by the combination of two different catalysts. We report that monometals, such as jagged Pt nanowires, can exhibit bifunctional characteristics owing to its complex surface morphology, where one site prefers electrochemical proton adsorption and another is responsible for activation, resulting in a 4-fold increase in the activity. We find that the conventional design guideline that the sites with a 0 eV Gibbs free energy of adsorption are optimal for HER does not hold under alkaline conditions, and rather, an energy between -0.2 and 0.0 eV is shown to be optimal. At the reaction temperatures, the high activity arises from low-coordination-number (≤7) Pt atoms exposed by the jagged surface. Our current demonstration raises an emerging prospect to understand highly complex kinetic phenomena on the nanoscale in full by implementing end-to-end multiscale strategies.

Structure-Based Synthesizability Prediction of Crystals Using Partially Supervised Learning.

Predicting the synthesizability of inorganic materials is one of the major challenges in accelerated material discovery. A widely employed approximate approach is to consider the thermodynamic decomposition stability due to its simplicity of computing, but it is notorious for either producing too many candidates or missing important metastable materials. These results, however, are not unexcepted since the synthesizability is a complex phenomenon, and the thermodynamic stability is just one contributor. Here, we suggest a machine-learning model to quantify the probability of synthesis based on the partially supervised learning of materials database. We adapted the positive and unlabeled machine learning (PU learning) by implementing the graph convolutional neural network as a classifier in which the model outputs crystal-likeness scores (CLscore). The model shows 87.4% true positive (CLscore > 0.5) prediction accuracy for the test set of experimentally reported cases (9356 materials) in the Materials Project. We further validated the model by predicting the synthesizability of newly reported experimental materials in the last 5 years (2015-2019) with an 86.2% true positive rate using the model trained with the database as of the end of year 2014. Our analysis shows that our model captures the structural motif for synthesizability beyond what is possible by Ehull. We find that 71 materials among the top 100 high-scoring virtual materials have indeed been previously synthesized in the literature. With the proposed data-driven metric of the crystal-likeness score, high-throughput virtual screenings and generative models can benefit significantly by effectively reducing the chemical space that needs to be explored experimentally in the future toward more rational materials design.

Uncertainty-Quantified Hybrid Machine Learning/Density Functional Theory High Throughput Screening Method for Crystals.

Computational high throughput screening (HTS) has emerged as a significant tool in material science to accelerate the discovery of new materials with target properties in recent years. However, despite many successful cases in which HTS led to the novel discovery, currently, the major bottleneck in HTS is a large computational cost of density functional theory (DFT) calculations that scale cubically with system size, limiting the chemical space that can be explored. The present work aims at addressing this computational burden of HTS by presenting a machine learning (ML) framework that can efficiently explore the chemical space. Our model is built upon an existing crystal graph convolutional neural network (CGCNN) to obtain formation energy of a crystal structure but is modified to allow uncertainty quantification for each prediction using the hyperbolic tangent activation function and dropout algorithm (CGCNN-HD). The uncertainty quantification is particularly important since typical usage of CGCNN (due to the lack of gradient implementation) does not involve structural relaxation which could cause substantial prediction errors. The proposed method is benchmarked against an existing application that identified promising photoanode material among the >7,000 hypothetical Mg-Mn-O ternary compounds using all DFT-HTS. In our approach, we perform the approximate HTS using CGCNN-HD and refine the results using full DFT for those selected (denoted as ML/DFT-HTS). The proposed hybrid model reduces the required DFT calculations by a factor of >50 compared to the previous DFT-HTS in making the same discovery of Mg2MnO4, experimentally validated new photoanode material. Further analysis demonstrates that the addition of HD components with uncertainty measures in the CGCNN-HD model increased the discoverability of promising materials relative to all DFT-HTS from 30% (CGCNN) to 68% (CGCNN-HD). The present ML/DFT-HTS with uncertainty quantification can thus be a fast alternative to DFT-HTS for efficient exploration of the vast chemical space.

Selective nitrogen capture by porous hybrid materials containing accessible transition metal ion sites.

Selective dinitrogen binding to transition metal ions mainly covers two strategic domains: biological nitrogen fixation catalysed by metalloenzyme nitrogenases, and adsorptive purification of natural gas and air. Many transition metal-dinitrogen complexes have been envisaged for biomimetic nitrogen fixation to produce ammonia. Inspired by this concept, here we report mesoporous metal-organic framework materials containing accessible Cr(III) sites, able to thermodynamically capture N2 over CH4 and O2. This fundamental study integrating advanced experimental and computational tools confirmed that the separation mechanism for both N2/CH4 and N2/O2 gas mixtures is driven by the presence of these unsaturated Cr(III) sites that allows a much stronger binding of N2 over the two other gases. Besides the potential breakthrough in adsorption-based technologies, this proof of concept could open new horizons to address several challenges in chemistry, including the design of heterogeneous biomimetic catalysts through nitrogen fixation.

Polyselenide Anchoring Using Transition-Metal Disulfides for Enhanced Lithium-Selenium Batteries.

While selenium has recently been proposed as a lithium battery cathode as a promising alternative to a lithium-sulfur battery, dissolution of intermediate species should be resolved to improve its cycle stability. Here, we report the promising results of transition-metal disulfides as an anchoring material and the underlying origin for preventing active material loss from the electrode using density functional theory calculations. Group 5 and 4 disulfides (VS2, NbS2, TaS2, TiS2, ZrS2, and HfS2) in particular show anchoring capabilities superior to those of group 6 disulfides (CrS2, MoS2, and WS2). The governing interaction controlling the latter relative anchoring strengths is shown to be charge transfer as understood by crystal-field theory. The current findings and methodologies provide novel chemical insight for the further design of inorganic anchoring materials for both lithium-selenium and lithium-sulfur batteries.

Computational exploration of borophane-supported single transition metal atoms as potential oxygen reduction and evolution electrocatalysts.

Novel monolayer-boron (borophene) is a recent addition to the family of 2D materials. In particular, full surface hydrogenation of triangular borophene (borophane (BH)) to passivate empty p orbitals in boron is identified as producing a new stable 2D material that possesses direction-dependent Dirac cones similar to graphene. By a series of density functional theory (DFT) computations, we investigated the potential of single transition metal atoms supported on borophane with vacancies (the TM-BH system) as an efficient ORR/OER electrocatalyst for applications in renewable energy technologies. In TM-BH systems, the coupling of d-orbitals of the TM dopant with the p-orbitals of surrounding boron atoms results in an increase in the density of states near the Fermi-level generating active sites to facilitate the ORR/OER via an efficient four-electron transfer mechanism. Among the considered TM-BH systems, Fe-BH and Rh-BH were found to be promising ORR electrocatalysts with overpotentials (ηORR) of 0.43 V and 0.47 V, respectively, whereas, for the OER, Rh-BH with 0.24 V has the smallest ηOER value followed by Co-BH (0.37 V), under the equilibrium electrode potential. These ηORR and ηOER values indicate higher activities than the current most active ORR (Pt(111) (0.63 V)) and OER (rutile-type RuO2 (0.37 V)) electrocatalysts.

A catalytic role of surface silanol groups in CO2 capture on the amine-anchored silica support.

A new mechanism of CO2 capture on the amine-functionalized silica support is demonstrated using density functional theory calculations, in which the silica surface not only acts as a support to anchor amines, but also can actively participate in the CO2 capture process through a facile proton transfer reaction with the amine groups. The surface-mediated proton transfer mechanism in forming a carbamate-ammonium product has lower kinetic barrier (8.1 kcal mol-1) than the generally accepted intermolecular mechanism (12.7 kcal mol-1) under dry conditions, and comparable to that of the water-assisted intermolecular mechanism (6.0 kcal mol-1) under humid conditions. These findings suggest that the CO2 adsorption on the amine-anchored silica surface would mostly occur via the rate-determining proton transfer step that is catalyzed by the surface silanol groups.

A local environment descriptor for machine-learned density functional theory at the generalized gradient approximation level.

We propose a grid-based local representation of electronic quantities that can be used in machine learning applications for molecules, which is compact, fixed in size, and able to distinguish different chemical environments. We apply the proposed approach to represent the external potential in density functional theory with modified pseudopotentials and demonstrate its proof of concept by predicting the Perdew-Burke-Ernzerhof and local density approximation electronic density and exchange-correlation potentials by kernel ridge regression. For 16 small molecules consisting of C, H, N, and O, the mean absolute error of exchange-correlation energy was 0.78 kcal/mol when trained for individual molecules. Furthermore, the model is shown to predict the exchange-correlation energy with an accuracy of 3.68 kcal/mol when the model is trained with a small fraction (4%) of all 16 molecules of the present dataset, suggesting a promising possibility that the current machine-learned model may predict the exchange-correlation energies of an arbitrary molecule with reasonable accuracy when trained with a sufficient amount of data covering an extensive variety of chemical environments.

Stability, Molecular Sieving, and Ion Diffusion Selectivity of a Lamellar Membrane from Two-Dimensional Molybdenum Disulfide.

Two-dimensional (2D) subnanometer channels allow unique mass transport promising for molecular sieving. New 2D channels of MoS2 nanosheets allow one to understand molecular transmission and separation, unlike the graphene oxide counterpart containing various defects and cationic metal contaminants. Membranes from layered MoS2 platelets show extraordinary stability in an aqueous environment and compatibility with polymer filters, both beneficial to efficient manufacturing. Sharing gas-tightness and unimpeded water vapor permeation with a graphene oxide membrane, our lamellar MoS2 membrane demonstrates a molecular sieving property for organic vapor for the first time. The MoS2 membrane also reveals diffusion selectivity of aqueous ions, attributable to the energy penalty in bulk-to-2D dimensional transition. These newly revealed properties of the lamellar membrane full of angstrom-sized 2D channels point to membrane technology applications for energy and environment.

Artificial neural network for the configuration problem in solids.

A machine learning approach based on the artificial neural network (ANN) is applied for the configuration problem in solids. The proposed method provides a direct mapping from configuration vectors to energies. The benchmark conducted for the M1 phase of Mo-V-Te-Nb oxide showed that only a fraction of configurations needs to be calculated, thus the computational burden significantly decreased, by a factor of 20-50, with R2 = 0.96 and MAD = 0.12 eV. It is shown that ANN can also handle the effects of geometry relaxation when properly trained, resulting in R2 = 0.95 and MAD = 0.13 eV.

Role of intermediate phase for stable cycling of Na7V4(P2O7)4PO4 in sodium ion battery.

Sodium ion batteries offer promising opportunities in emerging utility grid applications because of the low cost of raw materials, yet low energy density and limited cycle life remain critical drawbacks in their electrochemical operations. Herein, we report a vanadium-based ortho-diphosphate, Na7V4(P2O7)4PO4, or VODP, that significantly reduces all these drawbacks. Indeed, VODP exhibits single-valued voltage plateaus at 3.88 V vs. Na/Na(+) while retaining substantial capacity (>78%) over 1,000 cycles. Electronic structure calculations reveal that the remarkable single plateau and cycle life originate from an intermediate phase (a very shallow voltage step) that is similar both in the energy level and lattice parameters to those of fully intercalated and deintercalated states. We propose a theoretical scheme in which the reaction barrier that arises from lattice mismatches can be evaluated by using a simple energetic consideration, suggesting that the presence of intermediate phases is beneficial for cell kinetics by buffering the differences in lattice parameters between initial and final phases. We expect these insights into the role of intermediate phases found for VODP hold in general and thus provide a helpful guideline in the further understanding and design of battery materials.

Alcohol Dimer is Requisite to Form an Alkyl Oxonium Ion in the Proton Transfer of a Strong (Photo)Acid to Alcohol.

Alcohols, the simplest amphiprotic organic compounds, can exhibit either acidic or basic behavior by donating or accepting a proton. In this study, proton dissociation of a model photoacid in solution is explored by using time-resolved spectroscopy, revealing quantitatively for the first time that alcohol acts as a Brønsted base because of H-bonded cluster formation to enhance the reactivity. The protonated alcohol cluster, the alkyl oxonium ion, can be regarded as a key reaction intermediate in the well-established alcohol dehydration reaction. This finding signifies, as in water, the cooperativity of protic solvent molecules to facilitate nonaqueous acid-base reactions.

Adsorption of Carbon Dioxide on Unsaturated Metal Sites in M2 (dobpdc) Frameworks with Exceptional Structural Stability and Relation between Lewis Acidity and Adsorption Enthalpy.

A series of metal-organic frameworks (MOFs) M2 (dobpdc) (M=Mn, Co, Ni, Zn; H4 dobpdc=4,4'-dihydroxy-1,1'-biphenyl-3,3'-dicarboxylic acid), with a highly dense arrangement of open metal sites along hexagonal channels were prepared by microwave-assisted or simple solvothermal reactions. The activated materials were structurally expanded when guest molecules including CO2 were introduced into the pores. The Lewis acidity of the open metal sites varied in the order MnZn, as confirmed by C=O stretching bands in the IR spectra, which are related to the CO2 adsorption enthalpy. DFT calculations revealed that the high CO2 binding affinity of transition-metal-based M2 (dobpdc) is primarily attributable to the favorable charge transfer from CO2 (oxygen lone pair acting as a Lewis base) to the open metal sites (Lewis acid), while electrostatic effects, the underlying factor responsible for the particular order of binding strength observed across different transition metals, also play a role. The framework stability against water coincides with the order of Lewis acidity. In this series of MOFs, the structural stability of Ni2 (dobpdc) is exceptional; it endured in water vapor, liquid water, and in refluxing water for one month, and the solid remained intact on exposure to solutions of pH 2-13. The DFT calculations also support the experimental finding that Ni2 (dobpdc) has higher chemical stability than the other frameworks.