Kernel Methods for Predicting Yields of Chemical Reactions

Abstract

The use of machine learning methods for the prediction of reaction yield is an emerging area. We demonstrate the applicability of support vector regression (SVR) for predicting reaction yields, using combinatorial data. Molecular descriptors used in regression tasks related to chemical reac?tivity have often been based on time-consuming, computationally demanding quantum chemical calculations, usually density functional theory. Structure-based descriptors (molecular fingerprints and molecular graphs) are quicker and easier to calculate, and are applicable to any molecule. In this study, SVR models built on structure-based descriptors were compared to models built on quantum chemical descriptors. The models were evaluated along the dimension of each reaction component in a set of Buchwald-Hartwig amination reactions. The structure-based SVR models out-performed the quantum chemical SVR models, along the dimension of each reaction compo?nent. The applicability of the models was assessed with respect to similarity to training. Prospec?tive predictions of unseen Buchwald-Hartwig reactions are presented for synthetic assessment, to validate the generalisability of the models, with particular interest along the aryl halide dimension.

Data-Driven Multi-Objective Optimization Tactics for Catalytic Asymmetric Reactions Using Bisphosphine Ligands

“…The concept of a domain of applicability is well-established in quantitative structure− activity relationship (QSAR) models 80 and has previously been reported in reaction informatics. 81 Generally, the uncertainty in a particular prediction is correlated to the similarity between that molecule and the molecules used to construct the model. Using Euclidean distance in the chemical space to quantify the domain of applicability (see the Supporting Information for details on distance calculation), ligands tested that were 0.5 normalized parameter space units or less from a training set ligand had a prediction mean absolute error (MAE) of 0.24 kcal/ mol for the Hayashi−Heck reaction (Figure 10, graph 2 and graph 3).…”

Machine intelligence for chemical reaction space

“…This model combined with data augmentation techniques showed improved robustness in the low‐data regime and allowed the estimation of the prediction uncertainty 134 . Haywood et al 135 further analyzed the different representations and compared molecular fingerprints with computed descriptors using support vector machine (SVM) and found fingerprints to give better results.…”

Machine learning for design principles for single atom catalysts towards electrochemical reactions

“…The hyperplane is completely dened by the data points that are closest to the plane and between the support vectors. SVM can also be used in mapping the non-separable data through the radial basis function (RBF) kernel by transforming a real space into a higher-dimensional space through several hyperplanes: 33,34 f ðxÞ ¼…”