Combining Machine Learning and Computational Chemistry for Predictive Insights Into Chemical Systems.

Machine learning models are poised to make a transformative impact on chemical sciences by dramatically accelerating computational algorithms and amplifying insights available from computational chemistry methods. However, achieving this requires a confluence and coaction of expertise in computer science and physical sciences. This Review is written for new and experienced researchers working at the intersection of both fields. We first provide concise tutorials of computational chemistry and machine learning methods, showing how insights involving both can be achieved. We follow with a critical review of noteworthy applications that demonstrate how computational chemistry and machine learning can be used together to provide insightful (and useful) predictions in molecular and materials modeling, retrosyntheses, catalysis, and drug design.

Quantum alchemy beyond singlets: Bonding in diatomic molecules with hydrogen.

Bonding energies play an essential role in describing the relative stability of molecules in chemical space. Therefore, methods employed to search chemical space need to capture the bonding behavior for a wide range of molecules, including radicals. In this work, we investigate the ability of quantum alchemy to capture the bonding behavior of hypothetical chemical compounds, specifically diatomic molecules involving hydrogen with various electronic structures. We evaluate equilibrium bond lengths, ionization energies, and electron affinities of these fundamental systems. We compare and contrast how well manual quantum alchemy calculations, i.e., quantum mechanics calculations in which the nuclear charge is altered, and quantum alchemy approximations using a Taylor series expansion can predict these molecular properties. Our results suggest that while manual quantum alchemy calculations outperform Taylor series approximations, truncations of Taylor series approximations after the second order provide the most accurate Taylor series predictions. Furthermore, these results suggest that trends in quantum alchemy predictions are generally dependent on the predicted property (i.e., equilibrium bond length, ionization energy, or electron affinity). Taken together, this work provides insight into how quantum alchemy predictions using a Taylor series expansion may be applied to future studies of non-singlet systems as well as the challenges that remain open for predicting the bonding behavior of such systems.

Evaluating quantum alchemy of atoms with thermodynamic cycles: Beyond ground electronic states.

Due to the sheer size of chemical and materials space, high-throughput computational screening thereof will require the development of new computational methods that are accurate, efficient, and transferable. These methods need to be applicable to electron configurations beyond ground states. To this end, we have systematically studied the applicability of quantum alchemy predictions using a Taylor series expansion on quantum mechanics (QM) calculations for single atoms with different electronic structures arising from different net charges and electron spin multiplicities. We first compare QM method accuracy to experimental quantities, including first and second ionization energies, electron affinities, and spin multiplet energy gaps, for a baseline understanding of QM reference data. Next, we investigate the intrinsic accuracy of "manual" quantum alchemy. This method uses QM calculations involving nuclear charge perturbations of one atom's basis set to model another. We then discuss the reliability of quantum alchemy based on Taylor series approximations at different orders of truncation. Overall, we find that the errors from finite basis set treatments in quantum alchemy are significantly reduced when thermodynamic cycles are employed, which highlights a route to improve quantum alchemy in explorations of chemical space. This work establishes important technical aspects that impact the accuracy of quantum alchemy predictions using a Taylor series and provides a foundation for further quantum alchemy studies.

Computational predictions of metal-macrocycle stability constants require accurate treatments of local solvent and pH effects.

Rational design of molecular chelating agents requires a detailed understanding of physicochemical ligand-metal interactions in solvent phase. Computational quantum chemistry methods should be able to provide this, but computational reports have shown poor accuracy when determining absolute binding constants for many chelating molecules. To understand why, we compare and benchmark static- and dynamics-based computational procedures for a range of monovalent and divalent cations binding to a conventional cryptand molecule: 2.2.2-cryptand ([2.2.2]). The benchmarking comparison shows that dynamics simulations using standard OPLS-AA classical potentials can reasonably predict binding constants for monovalent cations, but these procedures fail for divalent cations. We also consider computationally efficient static procedure using Kohn-Sham density functional theory (DFT) and cluster-continuum modeling that accounts for local microsolvation and pH effects. This approach accurately predicts binding energies for monovalent and divalent cations with an average error of 3.2 kcal mol-1 compared to experiment. This static procedure thus should be useful for future molecular screening efforts, and high absolute errors in the literature may be due to inadequate modeling of local solvent and pH effects.

Quantifying Uncertainties in Solvation Procedures for Modeling Aqueous Phase Reaction Mechanisms.

Computational quantum chemistry provides fundamental chemical and physical insights into solvated reaction mechanisms across many areas of chemistry, especially in homogeneous and heterogeneous renewable energy catalysis. Such reactions may depend on explicit interactions with ions and solvent molecules that are nontrivial to characterize. Rigorously modeling explicit solvent effects with molecular dynamics usually brings steep computational costs while the performance of continuum solvent models such as polarizable continuum model (PCM), charge-asymmetric nonlocally determined local-electric (CANDLE), conductor-like screening model for real solvents (COSMO-RS), and effective screening medium method with the reference interaction site model (ESM-RISM) are less well understood for reaction mechanisms. Here, we revisit a fundamental aqueous hydride transfer reaction-carbon dioxide (CO2) reduction by sodium borohydride (NaBH4)-as a test case to evaluate how different solvent models perform in aqueous phase charge migrations that would be relevant to renewable energy catalysis mechanisms. For this system, quantum mechanics/molecular mechanics (QM/MM) molecular dynamics simulations almost exactly reproduced energy profiles from QM simulations, and the Na+ counterion in the QM/MM simulations plays an insignificant role over ensemble averaged trajectories that describe the reaction pathway. However, solvent models used on static calculations gave much more variability in data depending on whether the system was modeled using explicit solvent shells and/or the counterion. We pinpoint this variability due to unphysical descriptions of charge-separated states in the gas phase (i.e., self-interaction errors), and we show that using more accurate hybrid functionals and/or explicit solvent shells lessens these errors. This work closes with recommended procedures for treating solvation in future computational efforts in studying renewable energy catalysis mechanisms.

Child lead exposure near abandoned lead acid battery recycling sites in a residential community in Bangladesh: Risk factors and the impact of soil remediation on blood lead levels.

Lead is a potent neurotoxin that is particularly detrimental to children's cognitive development. Batteries account for at least 80% of global lead use and unsafe battery recycling is a major contributor to childhood lead poisoning. Our objectives were to assess the intensity and nature of child lead exposure at abandoned, informal used lead acid battery (ULAB) recycling sites in Kathgora, Savar, Bangladesh, as well as to assess the feasibility and effectiveness of a soil remediation effort to reduce exposure. ULAB recycling operations were abandoned in 2016 due to complaints from residents, but the lead contamination remained in the soil after operations ceased. We measured soil and blood lead levels (BLLs) among 69 children living within 200 m of the ULAB recycling site once before, and twice after (7 and 14 months after), a multi-part remediation intervention involving soil capping, household cleaning, and awareness-raising activities. Due to attrition, the sample size of children decreased from 69 to 47 children at the 7-month post-intervention assessment and further to 25 children at 14 months. We conducted non-parametric tests to assess changes in soil lead levels and BLLs. We conducted baseline surveys, as well as semi-structured interviews and observations with residents throughout the study period to characterize exposure behaviors and the community perceptions. We conducted bivariate and multivariate regression analyses of exposure characteristics to determine the strongest predictors of baseline child BLLs. Prior to remediation, median soil lead concentrations were 1400 mg/kg, with a maximum of 119,000 mg/kg and dropped to a median of 55 mg/kg after remediation (p < 0.0001). Among the 47 children with both baseline and post-intervention time 1 measurements, BLLs dropped from a median of 21.3 µg/dL to 17.0 µg/dL at 7 months (p < 0.0001). Among the 25 children with all three measurements, BLLs dropped from a median of 22.6 µg/dL to 14.8 µg/dL after 14 months (p < 0.0001). At baseline, distance from a child's residence to the nearest abandoned ULAB site was the strongest predictor of BLLs and baseline BLLs were 31% higher for children living within 50 m from the sites compared to those living further away (n = 69, p = 0.028). Women and children spent time in the contaminated site daily and relied on it for their livelihoods and for recreation. Overall, this study highlights the intensity of lead exposure associated with the ULAB recycling industry. Additionally, we document the feasibility and effectiveness of a multi-part remediation intervention at a contaminated site embedded within a residential community; substantially reducing child BLLs and soil lead concentrations.

Heteroarylureas with fused bicyclic diamine cores as inhibitors of fatty acid amide hydrolase.

A series of mechanism-based heteroaryl urea fatty acid amide hydrolase (FAAH) inhibitors with fused bicyclic diamine cores is described. In contrast to compounds built around a piperazine core, most of the fused bicyclic diamine bearing analogs prepared exhibited greater potency against rFAAH than the human enzyme. Several compounds equipotent against both species were identified and profiled in vivo.

First-principles modeling of chemistry in mixed solvents: Where to go from here?

Mixed solvents (i.e., binary or higher order mixtures of ionic or nonionic liquids) play crucial roles in chemical syntheses, separations, and electrochemical devices because they can be tuned for specific reactions and applications. Apart from fully explicit solvation treatments that can be difficult to parameterize or computationally expensive, there is currently no well-established first-principles regimen for reliably modeling atomic-scale chemistry in mixed solvent environments. We offer our perspective on how this process could be achieved in the near future as mixed solvent systems become more explored using theoretical and computational chemistry. We first outline what makes mixed solvent systems far more complex compared to single-component solvents. An overview of current and promising techniques for modeling mixed solvent environments is provided. We focus on so-called hybrid solvation treatments such as the conductor-like screening model for real solvents and the reference interaction site model, which are far less computationally demanding than explicit simulations. We also propose that cluster-continuum approaches rooted in physically rigorous quasi-chemical theory provide a robust, yet practical, route for studying chemical processes in mixed solvents.

Thermodynamic Hydricities of Biomimetic Organic Hydride Donors.

Thermodynamic hydricities (Δ GH-) in acetonitrile and dimethyl sulfoxide have been calculated and experimentally measured for several metal-free hydride donors: NADH analogs (BNAH, CN-BNAH, Me-MNAH, HEH), methylene tetrahydromethanopterin analogs (BIMH, CAFH), acridine derivatives (Ph-AcrH, Me2N-AcrH, T-AcrH, 4OH, 2OH, 3NH), and a triarylmethane derivative (6OH). The calculated hydricity values, obtained using density functional theory, showed a reasonably good match (within 3 kcal/mol) with the experimental values, obtained using "potential p Ka" and "hydride-transfer" methods. The hydride donor abilities of model compounds were in the 48.7-85.8 kcal/mol (acetonitrile) and 46.9-84.1 kcal/mol (DMSO) range, making them comparable to previously studied first-row transition metal hydride complexes. To evaluate the relevance of entropic contribution to the overall hydricity, Gibbs free energy differences (Δ GH-) obtained in this work were compared with the enthalpy (Δ HH-) values obtained by others. The results indicate that, even though Δ HH- values exhibit the same trends as Δ GH-, the differences between room-temperature Δ GH- and Δ HH- values range from 3 to 9 kcal/mol. This study also reports a new metal-free hydride donor, namely, an acridine-based compound 3NH, whose hydricity exceeds that of NaBH4. Collectively, this work gives a perspective of use metal-free hydride catalysts in fuel-forming and other reduction processes.

Improving human health outcomes with a low-cost intervention to reduce exposures from lead acid battery recycling: Dong Mai, Vietnam.

This study details the first comprehensive evaluation of the efficacy of a soil lead mitigation project in Dong Mai village, Vietnam. The village's population had been subject to severe lead poisoning for at least a decade as a result of informal Used Lead Acid Battery (ULAB) recycling. Between July 2013 to February 2015, Pure Earth and the Centre for Environment and Community Development (Hanoi, Vietnam) implemented a multi-faceted environmental and human health intervention. The intervention consisted of a series of institutional and low-cost engineering controls including the capping of lead contaminated surface soils, cleaning of home interiors, an education campaign and the construction of a work-clothes changing and bathing facility. The mitigation project resulted in substantial declines in human and environmental lead levels. Remediated home yard and garden areas decreased from an average surface soil concentration of 3940mg/kg to <100mg/kg. One year after the intervention, blood lead levels in children (<6 years old) were reduced by an average of 67%-from a median of 40.4µg/dL to 13.3µg/dL. The Dong Mai project resulted in significantly decreased environmental and biological lead levels demonstrating that low-cost, rapid and well-coordinated interventions could be readily applied elsewhere to significantly reduce preventable human health harm.

A Single Amino Acid Difference between Mouse and Human 5-Lipoxygenase Activating Protein (FLAP) Explains the Speciation and Differential Pharmacology of Novel FLAP Inhibitors.

5-Lipoxygenase activating protein (FLAP) plays a critical role in the metabolism of arachidonic acid to leukotriene A4, the precursor to the potent pro-inflammatory mediators leukotriene B4 and leukotriene C4 Studies with small molecule inhibitors of FLAP have led to the discovery of a drug binding pocket on the protein surface, and several pharmaceutical companies have developed compounds and performed clinical trials. Crystallographic studies and mutational analyses have contributed to a general understanding of compound binding modes. During our own efforts, we identified two unique chemical series. One series demonstrated strong inhibition of human FLAP but differential pharmacology across species and was completely inactive in assays with mouse or rat FLAP. The other series was active across rodent FLAP, as well as human and dog FLAP. Comparison of rodent and human FLAP amino acid sequences together with an analysis of a published crystal structure led to the identification of amino acid residue 24 in the floor of the putative binding pocket as a likely candidate for the observed speciation. On that basis, we tested compounds for binding to human G24A and mouse A24G FLAP mutant variants and compared the data to that generated for wild type human and mouse FLAP. These studies confirmed that a single amino acid mutation was sufficient to reverse the speciation observed in wild type FLAP. In addition, a PK/PD method was established in canines to enable preclinical profiling of mouse-inactive compounds.

Quantum Chemical Analyses of BH4- and BH3 OH- Hydride Transfers to CO2 in Aqueous Solution with Potentials of Mean Force.

Biomimetic hydride transfer catalysts are a promising route to efficiently convert CO2 into more useful products, but a lack of understanding about their atomic-scale reaction mechanisms slows their development. To this end, we report a computational quantum chemistry study of how aqueous solvation influences CO2 reduction reactions facilitated by sodium borohydride (NaBH4 ) and a partially oxidized derivative (NaBH3 OH). This work compares 0 K reaction barriers from nudged elastic band calculations to free-energy barriers obtained at 300 K using potentials of mean force from umbrella sampling simulations. We show that explicitly treating free energies from reaction pathway sampling has anywhere from a small to a large effect on the reaction-energy profiles for aqueous-phase hydride transfers to CO2 . Sampling along predefined reaction coordinates is thus recommended when it is computationally feasible.

The SAR of brain penetration for a series of heteroaryl urea FAAH inhibitors.

The SAR of brain penetration for a series of heteroaryl piperazinyl- and piperadinyl-urea fatty acid amide hydrolase (FAAH) inhibitors is described. Brain/plasma (B/P) ratios ranging from >4:1 to as low as 0.02:1 were obtained through relatively simple structural changes to various regions of the heteroaryl urea scaffold. It was not possible to predict the degree of central nervous system (CNS) penetration from the volumes of distribution (Vd) obtained from pharmacokinetic (PK) experiments as very high Vds did not correlate with high B/P ratios. Similarly, calculated topological polar surface areas (TPSAs) did not consistently correlate with the degree of brain penetration. The lowest B/P ratios were observed for those compounds that were significantly ionized at physiological pH. However, as this class of compounds inhibits the FAAH enzyme through covalent modification, low B/P ratios did not preclude effective central target engagement.

Structural and Substituent Group Effects on Multielectron Standard Reduction Potentials of Aromatic N-Heterocycles.

Aromatic N-heterocycles have been used in electrochemical CO2 reduction, but their precise role is not yet fully understood. We used first-principles quantum chemistry to determine how the molecular sizes and substituent groups of these molecules affect their standard redox potentials involving various proton and electron transfers. We then use that data to generate molecular Pourbaix diagrams to find the electrochemical conditions at which the aromatic N-heterocycle molecules could participate in multiproton and electron shuttling in accordance with the Sabatier principle. While one-electron standard redox potentials for aromatic N-heterocycles can vary significantly with molecule size and the presence of substituent groups, the two-electron and two-proton standard redox potentials depend much less on structural modifications and substituent groups. This indicates that a wide variety of aromatic N-heterocycles can participate in proton, electron, and/or hydride shuttling under suitable electrochemical conditions.

Oxygen transport in perovskite-type solid oxide fuel cell materials: insights from quantum mechanics.

CONSPECTUS: Global advances in industrialization are precipitating increasingly rapid consumption of fossil fuel resources and heightened levels of atmospheric CO2. World sustainability requires viable sources of renewable energy and its efficient use. First-principles quantum mechanics (QM) studies can help guide developments in energy technologies by characterizing complex material properties and predicting reaction mechanisms at the atomic scale. QM can provide unbiased, qualitative guidelines for experimentally tailoring materials for energy applications. This Account primarily reviews our recent QM studies of electrode materials for solid oxide fuel cells (SOFCs), a promising technology for clean, efficient power generation. SOFCs presently must operate at very high temperatures to allow transport of oxygen ions and electrons through solid-state electrolytes and electrodes. High temperatures, however, engender slow startup times and accelerate material degradation. SOFC technologies need cathode and anode materials that function well at lower temperatures, which have been realized with mixed ion-electron conductor (MIEC) materials. Unfortunately, the complexity of MIECs has inhibited the rational tailoring of improved SOFC materials. Here, we gather theoretically obtained insights into oxygen ion conductivity in two classes of perovskite-type materials for SOFC applications: the conventional La1-xSrxMO3 family (M = Cr, Mn, Fe, Co) and the new, promising class of Sr2Fe2-xMoxO6 materials. Using density functional theory + U (DFT+U) with U-J values obtained from ab initio theory, we have characterized the accompanying electronic structures for the two processes that govern ionic diffusion in these materials: (i) oxygen vacancy formation and (ii) vacancy-mediated oxygen migration. We show how the corresponding macroscopic oxygen diffusion coefficient can be accurately obtained in terms of microscopic quantities calculated with first-principles QM. We find that the oxygen vacancy formation energy is a robust descriptor for evaluating oxide ion transport properties. We also find it has a direct relationship with (i) the transition metal-oxygen bond strength and (ii) the extent to which electrons left behind by the departing oxygen delocalize onto the oxygen sublattice. Design principles from our QM results may guide further development of perovskite-based MIEC materials for SOFC applications.

Heteroarylureas with spirocyclic diamine cores as inhibitors of fatty acid amide hydrolase.

A series of mechanism based heteroaryl urea fatty acid amide hydrolase (FAAH) inhibitors with spirocyclic diamine cores is described. A potent member of this class, (37), was found to inhibit FAAH centrally, elevate the brain levels of three fatty acid ethanolamides [FAAs: anandamide (AEA), oleoyl ethanolamide (OEA) and palmitoyl ethanolamide (PEA)], and was moderately efficacious in a rat model of neuropathic pain.

1-Aryl-2-((6-aryl)pyrimidin-4-yl)amino)ethanols as competitive inhibitors of fatty acid amide hydrolase.

A series of 1-aryl-2-(((6-aryl)pyrimidin-4-yl)amino)ethanols have been found to be competitive inhibitors of fatty acid amide hydrolase (FAAH). One member of this class, JNJ-40413269, was found to have excellent pharmacokinetic properties, demonstrated robust central target engagement, and was efficacious in a rat model of neuropathic pain.

Accurate bond energies of biodiesel methyl esters from multireference averaged coupled-pair functional calculations.

Accurate bond dissociation energies (BDEs) are important for characterizing combustion chemistry, particularly the initial stages of pyrolysis. Here we contribute to evaluating the thermochemistry of biodiesel methyl ester molecules using ab initio BDEs derived from a multireference averaged coupled-pair functional (MRACPF2)-based scheme. Having previously validated this approach for hydrocarbons and a variety of oxygenates, herein we provide further validation for bonds within carboxylic acids and methyl esters, finding our scheme predicts BDEs within chemical accuracy (i.e., within 1 kcal/mol) for these molecules. Insights into BDE trends with ester size are then analyzed for methyl formate through methyl crotonate. We find that the carbonyl group in the ester moiety has only a local effect on BDEs. C═C double bonds in ester alkyl chains are found to increase the strengths of bonds adjacent to the double bond. An important exception are bonds beta to C═C or C═O bonds, which produce allylic-like radicals upon dissociation. The observed trends arise from different degrees of geometric relaxation and resonance stabilization in the radicals produced. We also compute BDEs in various small alkanes and alkenes as models for the long hydrocarbon chain of actual biodiesel methyl esters. We again show that allylic bonds in the alkenes are much weaker than those in the small methyl esters, indicating that hydrogen abstractions are more likely at the allylic site and even more likely at bis-allylic sites of alkyl chains due to more electrons involved in π-resonance in the latter. Lastly, we use the BDEs in small surrogates to estimate heretofore unknown BDEs in large methyl esters of biodiesel fuels.

Trends in bond dissociation energies of alcohols and aldehydes computed with multireference averaged coupled-pair functional theory.

As part of our ongoing investigation of the combustion chemistry of oxygenated molecules using multireference correlated wave function methods, we report bond dissociation energies (BDEs) in C1-C4 alcohols (from methanol to the four isomers of butanol) and C1-C4 aldehydes (from methanal to butanal). The BDEs are calculated with a multireference averaged coupled-pair functional-based scheme. We compare these multireference BDEs with those derived from experiment and single-reference methods. Trends in BDEs for the alcohols and aldehydes are rationalized by considering geometry relaxations of dissociated radical fragments, resonance stabilization, and hyperconjugation. Lastly, we discuss the conjectured association between bond strengths and rates of hydrogen abstraction by hydroxyl radicals. In general, abstraction reaction rates are higher at sites where the C-H bond energies are lower (and vice versa). However, comparison with available rate data shows this inverse relationship between bond strengths and abstraction rates does not hold at all temperatures.