Importance of Engineered and Learned Molecular Representations in Predicting Organic Reactivity, Selectivity, and Chemical Properties

Abstract

Conspectus Machine-readable chemical structure representations are foundational in all attempts to harness machine learning for the prediction of reactivities, selectivities, and chemical properties directly from molecular structure. The featurization of discrete chemical structures into a continuous vector space is a critical phase undertaken before model selection, and the development of new ways to quantitatively encode molecules is an active area of research. In this Account, we highlight the application and suitability of different representations, from expert-guided “engineered” descriptors to automatically “learned” features, in different prediction tasks relevant to organic and organometallic chemistry, where differing amounts of training data are available. These tasks include statistical models of stereo- and enantioselectivity, thermochemistry, and kinetics developed using experimental and quantum chemical data. The use of expert-guided molecular descriptors provides an opportunity to incorporate chemical knowledge, domain expertise, and physical constraints into statistical modeling. In applications to stereoselective organic and organometallic catalysis, where data sets may be relatively small and 3D-geometries and conformations play an important role, mechanistically informed features can be used successfully to obtain predictive statistical models that are also chemically interpretable. We provide an overview of several recent applications of this approach to obtain quantitative models for reactivity and selectivity, where topological descriptors, quantum mechanical calculations of electronic and steric properties, along with conformational ensembles, all feature as essential ingredients of the molecular representations used. Alternatively, more flexible, general-purpose molecular representations such as attributed molecular graphs can be used with machine learning approaches to learn the complex relationship between a structure and prediction target. This approach has the potential to out-perform more traditional representation methods such as “hand-crafted” molecular descriptors, particularly as data set sizes grow. One area where this is particularly relevant is in the use of large sets of quantum mechanical data to train quantitative structure–property relationships. A general approach toward curating useful data sets and training highly accurate graph neural network models is discussed in the context of organic bond dissociation enthalpies, where this strategy outperforms regression using precomputed descriptors. Finally, we describe how graph neural network predictions can be incorporated into mechanistically informed statistical models of chemical reactivity and selectivity. Once trained, this approach avoids the expensive computational overhead associated with quantum mechanical calculations, while maintaining chemical interpretability. We illustrate examples for which fast predictions of bond dissociation enthalpy and of the identities of radicals formed through cleavage of a molecule’s we...

Melding of Experiment and Theory Illuminates Mechanisms of Metal-Catalyzed Rearrangements: Computational Approaches and Caveats

“…188,189 Others are harnessing the power of machine learning methods for accelerated reaction discovery and chemical space exploration. [190][191][192] Nonetheless, we hope that we have given readers a snapshot of the utility of computational approaches through tales of transition-metalcatalyzed sigmatropic rearrangements. Timely publications describing experimental and computational aspects related to this topic have come out during the preparation of this review and we hope that interested readers will check them out for further reading.…”

Computational Organometallic Catalysis: Where We Are, Where We Are Going

“…A number of very recent reviews prove the fact that they are called to change the way in which catalysts are discovered. [51][52][53][54][55] The starting point of data-driven tools is establishing a relationship between a quantitative description of reactants, catalysts and reaction conditions with a property (for instance, activity: Quantitative Structure-Activity relationship, QSAR). The key ingredients of these models are the quantities (descriptors) that are correlated with the properties of interest.…”

“…Although these techniques are still at an early stage in this field, they have already achieved impressive successes, particularly in the area of enantioselective catalysis, [51,52] proving its potential. A number of very recent reviews prove the fact that they are called to change the way in which catalysts are discovered [51–55] …”

Real-time Prediction of 1H and 13C Chemical Shifts with DFT accuracy using a 3D Graph Neural Network

“…[23][24][25][26][27][28] When trained against a large number of experimentally measured chemical shifts, these methods have achieved predictive accuracies of 1.7 ppm for 13 C chemical shifts and 0.2 ppm for 1 H shifts (expressed as mean absolute error, MAE). 23 These earlier ML approaches tend to rely upon feature engineering 29 : expertcrafted rules are required to encode atomic environment, which can suffer from human bias and incompleteness, and which are often trained separately for different atom types (e.g., different models are developed for tetrahedral and trigonal carbon atoms). In particular, the rise of feature learning, as embodied by graph neural networks (GNNs), 30 has enabled 'end-to-end' learning from molecular structures and avoids rule-based encoding.…”