Machine Learning to Predict Diels-Alder Reaction Barriers from the Reactant State Electron Density

Data-Efficient, Chemistry-Aware Machine Learning Predictions of Diels-Alder Reaction Outcomes

The application of machine learning models to the prediction of reaction outcomes currently needs large and/or highly featurized data sets. We show that a chemistry-aware model, NERF, which mimics the bonding changes that occur during reactions, allows for highly accurate predictions of the outcomes of Diels-Alder reactions using a relatively small training set, with no pretraining and no additional features. We establish a diverse data set of 9537 intramolecular, hetero-, aromatic, and inverse electron demand Diels-Alder reactions. This data set is used to train a NERF model, and the performance is compared against state-of-the-art classification and generative machine learning models across low- and high-data regimes, with and without pretraining. The predictive accuracy (regio- and site selectivity in the major product) achieved by NERF exceeds 90% when as little as 40% of the data set is used for training. Another high-performing model, Chemformer, requires a larger training data set (>45%) and pretraining to reach 90% Top-1 accuracy. Accurate predictions of less-represented reaction subclasses, such as those involving heteroatomic or aromatic substrates, require higher percentages of training data. We also show how NERF can use small amounts of additional training data to quickly learn new systems and improve its overall understanding of reactivity. Synthetic chemists stand to benefit as this model can be rapidly expanded and tailored to areas of chemistry corresponding to the low-data regime.

Substituting density functional theory in reaction barrier calculations for hydrogen atom transfer in proteins

Hydrogen atom transfer (HAT) reactions are important in many biological systems. As these reactions are hard to observe experimentally, it is of high interest to shed light on them using simulations. Here, we present a machine learning model based on graph neural networks for the prediction of energy barriers of HAT reactions in proteins. As input, the model uses exclusively non-optimized structures as obtained from classical simulations. It was trained on more than 17 000 energy barriers calculated using hybrid density functional theory. We built and evaluated the model in the context of HAT in collagen, but we show that the same workflow can easily be applied to HAT reactions in other biological or synthetic polymers. We obtain for relevant reactions (small reaction distances) a model with good predictive power (R2 ∼ 0.9 and mean absolute error of <3 kcal mol-1). As the inference speed is high, this model enables evaluations of dozens of chemical situations within seconds. When combined with molecular dynamics in a kinetic Monte-Carlo scheme, the model paves the way toward reactive simulations.

Large Computational Survey of Intrinsic Reactivity of Aromatic Carbon Atoms with Respect to a Model Aldehyde Oxidase

Aldehyde oxidase (AOX) and other related molybdenum-containing enzymes are known to oxidize the C-H bonds of aromatic rings. This process contributes to the metabolism of pharmaceutical compounds and, therefore, is of vital importance to drug pharmacokinetics. The present work describes an automated computational workflow and its use for the prediction of intrinsic reactivity of small aromatic molecules toward a minimal model of the active site of AOX. The workflow is based on quantum chemical transition state searches for the underlying single-step oxidation reaction, where the automated protocol includes identification of unique aromatic C-H bonds, creation of three-dimensional reactant and product complex geometries via a templating approach, search for a transition state, and validation of reaction end points. Conformational search on the reactants, products, and the transition states is performed. The automated procedure has been validated on previously reported transition state barriers and was used to evaluate the intrinsic reactivity of nearly three hundred heterocycles commonly found in approved drug molecules. The intrinsic reactivity of more than 1000 individual aromatic carbon sites is reported. Stereochemical and conformational aspects of the oxidation reaction, which have not been discussed in previous studies, are shown to play important roles in accurate modeling of the oxidation reaction. Observations on structural trends that determine the reactivity are provided and rationalized.

Implementation of machine learning protocols to predict the hydrolysis reaction properties of organophosphorus substrates using descriptors of electron density topology

Prediction of catalytic reaction efficiency is one of the most intriguing and challenging applications of machine learning (ML) algorithms in chemistry. In this study, we demonstrated a strategy for utilizing ML protocols applied to Quantum Theory of Atoms In Molecules (QTAIM) parameters to predict the ability of the A17 L47K catalytic antibody to covalently capture organophosphate pesticides. We found that the novel "composite" DFT functional B97-3c could be effectively employed for fast and accurate initial geometry optimization, aligning well with the input dataset creation. QTAIM descriptors proved to be well-established in describing the examined dataset using density-based and hierarchical clustering algorithms. The obtained clusters exhibited correlations with the chemical classes of the input compounds. The precise physical interpretation of the QTAIM properties simplifies the explanation of feature impact for both supervised and unsupervised ML protocols. It also enables acceleration in the search for entries with desired properties within large databases. Furthermore, our findings indicated that Ridge Regression with Laplacian kernel and CatBoost Regressor algorithms demonstrated suitable performance in handling small datasets with non-trivial dependencies. They were able to predict the actual reaction barrier values with a high level of accuracy. Additionally, the CatBoost Classifier proved reliable in discriminating between "active" and "inactive" compounds.

Barrier Height Prediction by Machine Learning Correction of Semiempirical Calculations

Different machine learning (ML) models are proposed in the present work to predict density functional theory-quality barrier heights (BHs) from semiempirical quantum mechanical (SQM) calculations. The ML models include a multitask deep neural network, gradient-boosted trees by means of the XGBoost interface, and Gaussian process regression. The obtained mean absolute errors are similar to those of previous models considering the same number of data points. The ML corrections proposed in this paper could be useful for rapid screening of the large reaction networks that appear in combustion chemistry or in astrochemistry. Finally, our results show that 70% of the features with the highest impact on model output are bespoke predictors. This custom-made set of predictors could be employed by future Δ-ML models to improve the quantitative prediction of other reaction properties.

Benchmark of general-purpose machine learning-based quantum mechanical method AIQM1 on reaction barrier heights

Artificial intelligence-enhanced quantum mechanical method 1 (AIQM1) is a general-purpose method that was shown to achieve high accuracy for many applications with a speed close to its baseline semiempirical quantum mechanical (SQM) method ODM2*. Here, we evaluate the hitherto unknown performance of out-of-the-box AIQM1 without any refitting for reaction barrier heights on eight datasets, including a total of ∼24 thousand reactions. This evaluation shows that AIQM1's accuracy strongly depends on the type of transition state and ranges from excellent for rotation barriers to poor for, e.g., pericyclic reactions. AIQM1 clearly outperforms its baseline ODM2* method and, even more so, a popular universal potential, ANI-1ccx. Overall, however, AIQM1 accuracy largely remains similar to SQM methods (and B3LYP/6-31G* for most reaction types) suggesting that it is desirable to focus on improving AIQM1 performance for barrier heights in the future. We also show that the built-in uncertainty quantification helps in identifying confident predictions. The accuracy of confident AIQM1 predictions is approaching the level of popular density functional theory methods for most reaction types. Encouragingly, AIQM1 is rather robust for transition state optimizations, even for the type of reactions it struggles with the most. Single-point calculations with high-level methods on AIQM1-optimized geometries can be used to significantly improve barrier heights, which cannot be said for its baseline ODM2* method.

Geometry of Charge Density as a Reporter on the Role of the Protein Scaffold in Enzymatic Catalysis: Electrostatic Preorganization and Beyond

Enzymes host active sites inside protein macromolecules, which have diverse, often incredibly complex, and atom-expensive structures. It is an outstanding question what the role of these expensive scaffolds might be in enzymatic catalysis. Answering this question is essential to both enzymology and the design of artificial enzymes with proficiencies that will match those of the best natural enzymes. Protein rigidifying the active site, contrasted with the dynamics and vibrational motion promoting the reaction, as well as long-range electrostatics (also known as electrostatic preorganization) were all proposed as central contributions of the scaffold to the catalysis. Here, we show that all these effects inevitably produce changes in the quantum mechanical electron density in the active site, which in turn defines the reactivity. The phenomena are therefore fundamentally inseparable. The geometry of the electron density-a scalar field characterized by a number of mathematical features such as critical points-is a rigorous and convenient descriptor of enzymatic catalysis and a reporter on the role of the protein. We show how this geometry can be analyzed, linked to the reaction barriers, and report in particular on intramolecular electric fields in enzymes. We illustrate these tools on the studies of electrostatic preorganization in several representative enzyme classes, both natural and artificial. We highlight the forward-looking aspects of the approach.

Unicorns, Rhinoceroses and Chemical Bonds

The nascent field of computationally aided molecular design will be built around the ability to make computation useful to synthetic chemists who draw on their empirically based chemical intuition to synthesize new and useful molecules. This fact poses a dilemma, as much of existing chemical intuition is framed in the language of chemical bonds, which are pictured as possessing physical properties. Unfortunately, it has been posited that calculating these bond properties is impossible because chemical bonds do not exist. For much of the computationalchemistry community, bonds are seen as mythical-the unicorns of the chemical world. Here, we show that this is not the case. Using the same formalism and concepts that illuminated the atoms in molecules, we shine light on the bonds that connect them. The real space analogue of the chemical bond becomes the bond bundle in an extended quantum theory of atoms in molecules (QTAIM). We show that bond bundles possess all the properties typically associated with chemical bonds, including an energy and electron count. In addition, bond bundles are characterized by a number of nontraditional attributes, including, significantly, a boundary. We show, with examples drawn from solid state and molecular chemistry, that the calculated properties of bond bundles are consistent with those that nourish chemical intuition. We go further, however, and show that bond bundles provide new and quantifiable insights into the structure and properties of molecules and materials.

Unveiling the structural features that regulate carbapenem deacylation in KPC-2 through QM/MM and interpretable machine learning

Resistance to carbapenem β-lactams presents major clinical and economical challenges for the treatment of pathogen infections. The fast hydrolysis of carbapenems by carbapenemase-producing bacterial strains enables the effective deactivation of carbapenem antibiotics. In this study, we aim to unravel the structural features that distinguish the notable deacylation activity of carbapenemases. The deacylation reactions between imipenem (IPM) and the KPC-2 class A serine-based β-lactamases (ASβLs) are modeled with combined quantum mechanical/molecular mechanical (QM/MM) minimum energy pathway (MEP) calculations and interpretable machine-learning (ML) methods. We first applied a dual-level computational protocol to achieve fast sampling of QM/MM MEPs. A tree-based ensemble ML model was employed to learn the MEP activation barriers from the conformational features of the KPC-2/IPM active site. The barrier-predicting model was then unboxed using the Shapley additive explanation (SHAP) importance attribution methods to derive mechanistic insights, which were also verified by additional QM/MM wavefunction analysis. Essentially, we show that potential hydrogen bonding interactions of the general base and the tautomerization states of the carbapenem pyrroline ring could concertedly regulate the activation barrier of KPC-2/IPM deacylation. Nonetheless, we demonstrate the efficacy of interpretable ML to assist the analysis of QM/MM simulation data for robust extraction of human-interpretable mechanistic insights.

Bond Bundle Analysis of Ketosteroid Isomerase

Bond bundle analysis is used to investigate enzymatic catalysis in the ketosteroid isomerase (KSI) active site. We identify the unique bonding regions in five KSI systems, including those exposed to applied oriented electric fields and those with amino acid mutations, and calculate the precise redistribution of electron density and other regional properties that accompanies either enhancement or inhibition of KSI catalytic activity. We find that catalytic enhancement results from promoting both inter- and intra-molecular electron density redistribution, between bond bundles and bond wedges within the KSI-docked substrate molecule, in the forward direction of the catalyzed reaction. Though the redistribution applies to both types of perturbed systems and is thus suggestive of a general catalytic role, we observe that bond properties (e.g., volume vs energy vs electron count) can respond independently and disproportionately depending on the type of perturbation. We conclude that the resulting catalytic enhancement/inhibition proceeds via different mechanisms, where some bond properties are utilized more by one type of perturbation than the other. Additionally, we find that the correlations between bond wedge properties and catalyzed reaction barrier energies are additive to predict those of bond bundles and atomic basins, providing a rigorous grounding for connecting changes in local charge density to resulting shifts in reaction barrier energy.

NNAIMQ: A neural network model for predicting QTAIM charges

Atomic charges provide crucial information about the electronic structure of a molecular system. Among the different definitions of these descriptors, the one proposed by the Quantum Theory of Atoms in Molecules (QTAIM) is particularly attractive given its invariance against orbital transformations although the computational cost associated with their calculation limits its applicability. Given that Machine Learning (ML) techniques have been shown to accelerate orders of magnitude the computation of a number of quantum mechanical observables, in this work, we take advantage of ML knowledge to develop an intuitive and fast neural network model (NNAIMQ) for the computation of QTAIM charges for C, H, O, and N atoms with high accuracy. Our model has been trained and tested using data from quantum chemical calculations in more than 45 000 molecular environments of the near-equilibrium CHON chemical space. The reliability and performance of NNAIMQ have been analyzed in a variety of scenarios, from equilibrium geometries to molecular dynamics simulations. Altogether, NNAIMQ yields remarkably small prediction errors, well below the 0.03 electron limit in the general case, while accelerating the calculation of QTAIM charges by several orders of magnitude.

New venues in electron density analysis

We provide a comprehensive overview of the chemical information within the electron density: how to extract information, but also how to obtain and how to assess the quality of the...