Applying Large Graph Neural Networks to Predict Transition Metal Complex Energies Using the tmQM_wB97MV Data Set.

Machine learning (ML) methods have shown promise for discovering novel catalysts but are often restricted to specific chemical domains. Generalizable ML models require large and diverse training data sets, which exist for heterogeneous catalysis but not for homogeneous catalysis. The tmQM data set, which contains properties of 86,665 transition metal complexes calculated at the TPSSh/def2-SVP level of density functional theory (DFT), provided a promising training data set for homogeneous catalyst systems. However, we find that ML models trained on tmQM consistently underpredict the energies of a chemically distinct subset of the data. To address this, we present the tmQM_wB97MV data set, which filters out several structures in tmQM found to be missing hydrogens and recomputes the energies of all other structures at the ωB97M-V/def2-SVPD level of DFT. ML models trained on tmQM_wB97MV show no pattern of consistently incorrect predictions and much lower errors than those trained on tmQM. The ML models tested on tmQM_wB97MV were, from best to worst, GemNet-T > PaiNN ≈ SpinConv > SchNet. Performance consistently improves when using only neutral structures instead of the entire data set. However, while models saturate with only neutral structures, more data continue to improve the models when including charged species, indicating the importance of accurately capturing a range of oxidation states in future data generation and model development. Furthermore, a fine-tuning approach in which weights were initialized from models trained on OC20 led to drastic improvements in model performance, indicating transferability between ML strategies of heterogeneous and homogeneous systems.

On scientific understanding with artificial intelligence.

An oracle that correctly predicts the outcome of every particle physics experiment, the products of every possible chemical reaction or the function of every protein would revolutionize science and technology. However, scientists would not be entirely satisfied because they would want to comprehend how the oracle made these predictions. This is scientific understanding, one of the main aims of science. With the increase in the available computational power and advances in artificial intelligence, a natural question arises: how can advanced computational systems, and specifically artificial intelligence, contribute to new scientific understanding or gain it autonomously? Trying to answer this question, we adopted a definition of 'scientific understanding' from the philosophy of science that enabled us to overview the scattered literature on the topic and, combined with dozens of anecdotes from scientists, map out three dimensions of computer-assisted scientific understanding. For each dimension, we review the existing state of the art and discuss future developments. We hope that this Perspective will inspire and focus research directions in this multidisciplinary emerging field.

A Comprehensive Discovery Platform for Organophosphorus Ligands for Catalysis.

The design of molecular catalysts typically involves reconciling multiple conflicting property requirements, largely relying on human intuition and local structural searches. However, the vast number of potential catalysts requires pruning of the candidate space by efficient property prediction with quantitative structure-property relationships. Data-driven workflows embedded in a library of potential catalysts can be used to build predictive models for catalyst performance and serve as a blueprint for novel catalyst designs. Herein we introduce kraken, a discovery platform covering monodentate organophosphorus(III) ligands providing comprehensive physicochemical descriptors based on representative conformer ensembles. Using quantum-mechanical methods, we calculated descriptors for 1558 ligands, including commercially available examples, and trained machine learning models to predict properties of over 300000 new ligands. We demonstrate the application of kraken to systematically explore the property space of organophosphorus ligands and how existing data sets in catalysis can be used to accelerate ligand selection during reaction optimization.

Mapping C-H⋠⋠⋠M Interactions in Confined Spaces: (α-ICyDMe )Au, Ag, Cu Complexes Reveal "Contra-electrostatic H Bonds" Masquerading as Anagostic Interactions*.

What happens when a C-H bond is forced to interact with unpaired pairs of electrons at a positively charged metal? Such interactions can be considered as "contra-electrostatic" H-bonds, which combine the familiar orbital interaction pattern characteristic for the covalent contribution to the conventional H-bonding with an unusual contra-electrostatic component. While electrostatics is strongly stabilizing component in the conventional C-H⋅⋅⋅X bonds where X is an electronegative main group element, it is destabilizing in the C-H⋅⋅⋅M contacts when M is Au(I), Ag(I), or Cu(I) of NHC-M-Cl systems. Such remarkable C-H⋅⋅⋅M interaction became experimentally accessible within (α-ICyDMe )MCl, NHC-Metal complexes embedded into cyclodextrins. Computational analysis of the model systems suggests that the overall interaction energies are relatively insensitive to moderate variations in the directionality of interaction between a C-H bond and the metal center, indicating stereoelectronic promiscuity of fully filled set of d-orbitals. A combination of experimental and computational data demonstrates that metal encapsulation inside the cyclodextrin cavity forces the C-H bond to point toward the metal, and reveals a still attractive "contra-electrostatic" H-bonding interaction.

Illuminating the dark conformational space of macrocycles using dominant rotors.

Three-dimensional conformation is the primary determinant of molecular properties. The thermal energy available at room temperature typically equilibrates the accessible conformational states. Here, we introduce a method for isolating unique and previously understudied conformations of macrocycles. The observation of unusual conformations of 16- to 22-membered rings has been made possible by controlling their interconversion using dominant rotors, which represent tunable atropisomeric constituents with relatively high rotational barriers. Density functional theory and in situ NMR measurements suggest that dominant rotor candidates for the amino-acid-based structures considered here should possess a rotational energy barrier of at least 25 kcal mol-1. Notable differences in the geometries of the macrocycle conformations were identified by NMR spectroscopy and X-ray crystallography. There is evidence that amino acid residues can be forced into rare turn motifs not observed in the corresponding linear counterparts and homodetic rings. These findings should unlock new avenues for studying the conformation-activity relationships of bioactive molecules.

Data-Driven Strategies for Accelerated Materials Design.

The ongoing revolution of the natural sciences by the advent of machine learning and artificial intelligence sparked significant interest in the material science community in recent years. The intrinsically high dimensionality of the space of realizable materials makes traditional approaches ineffective for large-scale explorations. Modern data science and machine learning tools developed for increasingly complicated problems are an attractive alternative. An imminent climate catastrophe calls for a clean energy transformation by overhauling current technologies within only several years of possible action available. Tackling this crisis requires the development of new materials at an unprecedented pace and scale. For example, organic photovoltaics have the potential to replace existing silicon-based materials to a large extent and open up new fields of application. In recent years, organic light-emitting diodes have emerged as state-of-the-art technology for digital screens and portable devices and are enabling new applications with flexible displays. Reticular frameworks allow the atom-precise synthesis of nanomaterials and promise to revolutionize the field by the potential to realize multifunctional nanoparticles with applications from gas storage, gas separation, and electrochemical energy storage to nanomedicine. In the recent decade, significant advances in all these fields have been facilitated by the comprehensive application of simulation and machine learning for property prediction, property optimization, and chemical space exploration enabled by considerable advances in computing power and algorithmic efficiency.In this Account, we review the most recent contributions of our group in this thriving field of machine learning for material science. We start with a summary of the most important material classes our group has been involved in, focusing on small molecules as organic electronic materials and crystalline materials. Specifically, we highlight the data-driven approaches we employed to speed up discovery and derive material design strategies. Subsequently, our focus lies on the data-driven methodologies our group has developed and employed, elaborating on high-throughput virtual screening, inverse molecular design, Bayesian optimization, and supervised learning. We discuss the general ideas, their working principles, and their use cases with examples of successful implementations in data-driven material discovery and design efforts. Furthermore, we elaborate on potential pitfalls and remaining challenges of these methods. Finally, we provide a brief outlook for the field as we foresee increasing adaptation and implementation of large scale data-driven approaches in material discovery and design campaigns.

Data-science driven autonomous process optimization.

Autonomous process optimization involves the human intervention-free exploration of a range process parameters to improve responses such as product yield and selectivity. Utilizing off-the-shelf components, we develop a closed-loop system for carrying out parallel autonomous process optimization experiments in batch. Upon implementation of our system in the optimization of a stereoselective Suzuki-Miyaura coupling, we find that the definition of a set of meaningful, broad, and unbiased process parameters is the most critical aspect of successful optimization. Importantly, we discern that phosphine ligand, a categorical parameter, is vital to determination of the reaction outcome. To date, categorical parameter selection has relied on chemical intuition, potentially introducing bias into the experimental design. In seeking a systematic method for selecting a diverse set of phosphine ligands, we develop a strategy that leverages computed molecular feature clustering. The resulting optimization uncovers conditions to selectively access the desired product isomer in high yield.

Machine learning dihydrogen activation in the chemical space surrounding Vaska's complex.

Homogeneous catalysis using transition metal complexes is ubiquitously used for organic synthesis, as well as technologically relevant in applications such as water splitting and CO2 reduction. The key steps underlying homogeneous catalysis require a specific combination of electronic and steric effects from the ligands bound to the metal center. Finding the optimal combination of ligands is a challenging task due to the exceedingly large number of possibilities and the non-trivial ligand-ligand interactions. The classic example of Vaska's complex, trans-[Ir(PPh3)2(CO)(Cl)], illustrates this scenario. The ligands of this species activate iridium for the oxidative addition of hydrogen, yielding the dihydride cis-[Ir(H)2(PPh3)2(CO)(Cl)] complex. Despite the simplicity of this system, thousands of derivatives can be formulated for the activation of H2, with a limited number of ligands belonging to the same general categories found in the original complex. In this work, we show how DFT and machine learning (ML) methods can be combined to enable the prediction of reactivity within large chemical spaces containing thousands of complexes. In a space of 2574 species derived from Vaska's complex, data from DFT calculations are used to train and test ML models that predict the H2-activation barrier. In contrast to experiments and calculations requiring several days to be completed, the ML models were trained and used on a laptop on a time-scale of minutes. As a first approach, we combined Bayesian-optimized artificial neural networks (ANN) with features derived from autocorrelation and deltametric functions. The resulting ANNs achieved high accuracies, with mean absolute errors (MAE) between 1 and 2 kcal mol-1, depending on the size of the training set. By using a Gaussian process (GP) model trained with a set of selected features, including fingerprints, accuracy was further enhanced. Remarkably, this GP model minimized the MAE below 1 kcal mol-1, by using only 20% or less of the data available for training. The gradient boosting (GB) method was also used to assess the relevance of the features, which was used for both feature selection and model interpretation purposes. Features accounting for chemical composition, atom size and electronegativity were found to be the most determinant in the predictions. Further, the ligand fragments with the strongest influence on the H2-activation barrier were identified.

The Cyclopropane Ring as a Reporter of Radical Leaving-Group Reactivity for Ni-Catalyzed C(sp3)-O Arylation.

The ability to understand and predict reactivity is essential for the development of new reactions. In the context of Ni-catalyzed C(sp3)-O functionalization, we have developed a unique strategy employing activated cyclopropanols to aid the design and optimization of a redox-active leaving group for C(sp3)-O arylation. In this chemistry, the cyclopropane ring acts as a reporter of leaving-group reactivity, since the ring-opened product is obtained under polar (2e) conditions, and the ring-closed product is obtained under radical (1e) conditions. Mechanistic studies demonstrate that the optimal leaving group is redox-active and are consistent with a Ni(I)/Ni(III) catalytic cycle. The optimized reaction conditions are also used to synthesize a number of arylcyclopropanes, which are valuable pharmaceutical motifs.

Twofold π-Extension of Polyarenes via Double and Triple Radical Alkyne peri-Annulations: Radical Cascades Converging on the Same Aromatic Core.

A versatile synthetic route to distannyl-substituted polyarenes was developed via double radical peri-annulations. The cyclization precursors were equipped with propargylic OMe traceless directing groups (TDGs) for regioselective Sn-radical attack at the triple bonds. The two peri-annulations converge at a variety of polycyclic cores to yield expanded difunctionalized polycyclic aromatic hydrocarbons (PAHs). This approach can be extended to triple peri-annulations, where annulations are coupled with a radical cascade that connects two preexisting aromatic cores via a formal C-H activation step. The installed Bu3Sn groups serve as chemical handles for further functionalization via direct cross-coupling, iodination, or protodestannylation and increase solubility of the products in organic solvents. Photophysical studies reveal that the Bu3Sn-substituted PAHs are moderately fluorescent, and their protodestannylation results in an up to 10-fold fluorescence quantum yield enhancement. DFT calculations identified the most likely possible mechanism of this complex chemical transformation involving two independent peri-cyclizations at the central core.

Testing the limits of radical-anionic CH-amination: a 10-million-fold decrease in basicity opens a new path to hydroxyisoindolines via a mixed C-N/C-O-forming cascade.

An intramolecular C(sp3)-H amidation proceeds in the presence of t-BuOK, molecular oxygen, and DMF. This transformation is initiated by the deprotonation of an acidic N-H bond and selective radical activation of a benzylic C-H bond towards hydrogen atom transfer (HAT). Cyclization of this radical-anion intermediate en route to a two-centered/three-electron (2c,3e) C-N bond removes electron density from nitrogen. As this electronegative element resists such an "oxidation", making nitrogen more electron rich is key to overcoming this problem. This work dramatically expands the range of N-anions that can participate in this process by using amides instead of anilines. The resulting 107-fold decrease in the N-component basicity (and nucleophilicity) doubles the activation barrier for C-N bond formation and makes this process nearly thermoneutral. Remarkably, this reaction also converts a weak reductant into a much stronger reductant. Such "reductant upconversion" allows mild oxidants like molecular oxygen to complete the first part of the cascade. In contrast, the second stage of NH/CH activation forms a highly stabilized radical-anion intermediate incapable of undergoing electron transfer to oxygen. Because the oxidation is unfavored, an alternative reaction path opens via coupling between the radical anion intermediate and either superoxide or hydroperoxide radical. The hydroperoxide intermediate transforms into the final hydroxyisoindoline products under basic conditions. The use of TEMPO as an additive was found to activate less reactive amides. The combination of experimental and computational data outlines a conceptually new mechanism for conversion of unprotected amides into hydroxyisoindolines proceeding as a sequence of C-H amidation and C-H oxidation.

Electronic Donation or Steric Contraction: A Spectroscopic and Structural Analysis of Medium-Sized Constrained Rings for Potential Long-Range Hyperconjugation.

Herein, we report the 1JCH analyses, natural bond orbital analyses, and X-ray crystal structures of a number of C, O, and N constrained tricyclic cycles. These experiments provide access into the nature of the apparent Perlin effect previously reported in constrained tricyclic cycles, as well as evidence suggesting both steric contraction and long-range hyperconjugation account for the observed 1JCH perturbations. We report a true Perlin effect of 10.9 Hz in an azocane and large steric effect resulting in Δ1JC-H = 10.9 Hz in a cyclooctane.

CO2 or SO2: Should It Stay, or Should It Go?

A broad computational analysis of carbon-centered radical formation via the loss of either CO2 or SO2 from the respective RXO2 radical precursors (X = C or S) reveals dramatic differences between these two types of dissociative processes. Whereas the C-C scission with the loss of CO2 is usually exothermic, the C-S scission with the loss of SO2 is generally endothermic. However, two factors can make the C-S scissions thermodynamically favorable: increased entropy, characteristic for the dissociative processes, and stereoelectronic influences of substituents. The threshold between endergonic and exergonic C-S fragmentations depends on subtle structural effects. In particular, the degree of fluorination in a radical precursor has a notable impact on the reaction outcome. This study aims to demystify the intricacies in reactivity regarding the generation of radicals from sulfinates and carboxylates, as related to their role in radical cross-coupling.

Stereoelectronic Influence of a "Spectator" Propargylic Substituent Can Override Aromaticity Effects in Radical Peri-Cyclizations en Route to Expanded Polyaromatics.

Computational analysis quantifies key trends in " peri"-radical cyclizations, a recently developed type of ring-forming reaction for the expansion of polyaromatic systems at the zigzag edge. Comparison of vinyl radical attack on the peri-position versus a topologically similar six-membered ring formation at the armchair edge reveals that the barriers for the peri-ring closure are slightly higher, even though the peri-attack is more exergonic. On the other hand, the intramolecular competition between the formation of a five-membered ring by ortho-attack at the armchair edge and formation of a six-membered ring by peri-attack at the zigzag edge clearly favors six-membered ring formation. The key novel finding is the unprecedented sensitivity of peri-cyclization to the presence and spatial orientation of a "spectator" propargylic -OMe substituent. Remarkably, formation of cis-products proceeds, in general, through a significantly (∼2-4 kcal/mol) lower barrier than formation of the trans-products, even when the cis-products are less stable. The origin of this unexpected effect is clearly stereoelectronic. These findings identify such remote substitution as a conceptually new tool for the control of rate and selectivity of radical reactions. The correlations of activation barriers for vinyl radical attack with aromaticity of the target show the expected relationship in phenanthrenes and pyrenes but not in anthracenes. In the latter case, the attack at the less aromatic ring corresponds to a higher barrier because a steric penalty on the stereoelectronically favorable cis-TS negates the accelerating influence of the properly aligned C-O and C-Sn bonds.

Interrupted Baeyer-Villiger Rearrangement: Building A Stereoelectronic Trap for the Criegee Intermediate.

The instability of hydroxy peroxyesters, the elusive Criegee intermediates of the Baeyer-Villiger rearrangement, can be alleviated by selective deactivation of the stereoelectronic effects that promote the 1,2-alkyl shift. Stable cyclic Criegee intermediates constrained within a five-membered ring can be prepared by mild reduction of the respective hydroperoxy peroxyesters (ß-hydroperoxy-ß-peroxylactones) which were formed in high yields in reaction of ß-ketoesters with BF3 ⋅Et2 O/H2 O2 .

Radical Alkyne peri-Annulation Reactions for the Synthesis of Functionalized Phenalenes, Benzanthrenes, and Olympicene.

Radical cyclization reactions at a peri position were used for the synthesis of polyaromatic compounds. Depending on the choice of reaction conditions and substrate, this flexible approach led to Bu3 Sn-substituted phenalene, benzanthrene, and olympicene derivatives. Subsequent reactions with electrophiles provided synthetic access to previously inaccessible functionalized polyaromatic compounds.

Stereoelectronic Interactions as a Probe for the Existence of the Intramolecular α-Effect.

The first systematic study of the intramolecular α-effect, both in the stable ground-state structures and in the high-energy intermediates, was accomplished using the anomeric effect as an internal stereoelectronic probe. Contrary to the expectations based on the simple orbital mixing model, the lone pairs in a pair of neutral directly connected heteroatoms are not raised in energy to become stronger donors toward adjacent σ- and π-acceptors. Instead, the key n(X-Y)→σ*C-F interactions (X,Y = O,N) in the "α-systems" (both acyclic and constrained within a heterocyclohexane frame) are weaker than nX→σ*C-F interactions in "normal" systems. Surprisingly, polar solvent effects increase the apparent magnitude of α-effect as measured via increase in the anomeric stabilization. This behavior is opposite to the solvent dependence of normal systems where the anomeric effect is severely weakened by polar solvents. This contrasting behavior reflects the different balance of electrostatic and conjugative interactions in the two types of anomeric systems: the α-systems suffer less from the unfavorable orientation of bond dipoles in the equatorial conformer, a destabilizing electrostatic effect that is shielded by the polar environments. A weak α-effect is brought to life when the buttressing α-heteroatom bears a negative charge. However, electrostatic components mask the role of stabilizing orbital interactions. In contrast, the increased electron demand in carbocations and related electron-deficient TS- like structures does not lead to activation of the α-effect. As a consequence, we observed that ethers are better radical- and cation-stabilizing groups than peroxides. The latter finding should have significant implications for understanding the mechanistic complexity associated with the interaction of carbonyl compounds with hydroperoxides and H2O2 in acidic media, as such reactions involve α-cationic intermediates.

Formaldehyde-Extruding Homolytic Aromatic Substitution via C→O Transposition: Selective 'Traceless-Linker' access to Congested Biaryl Bonds.

A new, selective way to form C-C bonds has been developed. In this report, we disclose the homolytic aromatic substitution via C→O transposition coupled with the elimination of formaldehyde (as a traceless linker). Computational analysis indicates the selectivity can be tuned by sterics in the starting materials following an ipso-attack that leads to the C→O transposition.

Stereoelectronic Control in the Ozone-Free Synthesis of Ozonides.

The value of stereoelectronic guidelines is illustrated by the discovery of a convenient, ozone-free synthesis of bridged secondary ozonides from 1,5-dicarbonyl compounds and H2 O2 . The tetraoxane products generally formed in reactions of carbonyl and dicarbonyl compounds with H2 O2 were not detected because the structural distortions imposed on the tetraoxacyclohexane subunit in [3.2.2]tetraoxanonanes by the three-carbon bridge leads to the partial deactivation of anomeric effects. The new procedure is readily scalable to produce gram quantities of the ozonides. This reaction enables the selective preparation of ozonides without the use of ozone.

Drawing Catalytic Power from Charge Separation: Stereoelectronic and Zwitterionic Assistance in the Au(I)-Catalyzed Bergman Cyclization.

The synergy between bond formation and bond breaking that is typical for pericyclic reactions is lost in their mechanistic cousins, cycloaromatization reactions. In these reactions, exemplified by the Bergman cyclization (BC), two bonds are sacrificed to form a single bond, and the reaction progress is interrupted at the stage of a cyclic diradical intermediate. The catalytic power of Au(I) in BC stems from a combination of two sources: stereoelectronic assistance of C-C bond formation (i.e., "LUMO umpolung") and crossover from a diradical to a zwitterionic mechanism that takes advantage of the catalyst's dual ability to stabilize both negative and positive charges. Not only does the synergy between the bond-forming and charge-delocalizing interactions lead to a dramatic (>hundred-billion-fold) acceleration, but the evolution of the two effects results in continuous reinforcement of the substrate/catalyst interaction along the cyclization path. This cooperativity converts the BC into the first example of an aborted [3,3] sigmatropic shift where the pericyclic "transition state" becomes the most stable species on the reaction hypersurface. Aborting the pericyclic path facilitates trapping of cyclic intermediate by a variety of further reactions and provides a foundation for the discovery of new modes of reactivity of polyunsaturated substrates. The application of distortion/interaction analysis allows us to quantify the increased affinity of Au-catalysts to the Bergman cyclization transition state as one of the key components of the large catalytic effect.