Scaffold‐Directed Face Selectivity Machine‐Learned from Vectors of Non‐covalent Interactions

Abstract

This work describes a method to vectorize and Machine-Learn, ML, non-covalent interactions responsible for scaffold-directed reactions important in synthetic chemistry. Models trained on this representation predict correct face of approach in ca. 90 % of Michael additions or Diels-Alder cycloadditions. These accuracies are significantly higher than those based on traditional ML descriptors, energetic calculations, or intuition of experienced synthetic chemists. Our results also emphasize the importance of ML models being provided with relevant mechanistic knowledge; without such knowledge, these models cannot easily "transfer-learn" and extrapolate to previously unseen reaction mechanisms.

Machine Learning May Sometimes Simply Capture Literature Popularity Trends: A Case Study of Heterocyclic Suzuki–Miyaura Coupling

“…This might be true when the descriptors used to construct the model capture the chemical essence of a problem in question, for example, when trying to predict reaction outcomes given its substrates, the structural, steric, and electronic descriptors are generally sufficient. [7][8][9][10]50 The condition prediction problem, however, is markedly different because in addition to the structural features of reactants, products, and reagents, it entails several "human" factors: Conditions are often chosen based on the query of relevant literature, ultimately selecting those most frequently reported (this may explain why popularity-based metrics worked nearly as well as ML). In addition, mundane factors of instantaneous availability of specific reagents/solvents in one's laboratory or even "historical" preference for certain choices (i.e., conditions commonly used in one's laboratory) might come into play.…”

RegioML: predicting the regioselectivity of electrophilic aromatic substitution reactions using machine learning

“…Many useful reactions are underutilised in synthetic organic chemistry because of an inability to predict the regioselectivity of the reaction, 1 and there is thus an increasing interest in developing regioselectivity prediction methods for such reactions. Recent examples include nucleophilic 2,3 and electrophilic aromatic substitution reactions, [4][5][6][7][8][9] Diels-Alder reactions, 10,11 Heck reactions, 12 radical C-H functionalisation of heterocycles, 13 and reactions such as alkylations, Michael additions, and aldol condensations that proceed through proton abstraction. 14 These methods have been based on either quantum chemical (QM) calculations, 2,5,6 machine learning (ML) trained on experimental data, 8,[10][11][12] or a combination of the two where QM has either provided descriptors for the ML model 3,9 or was used to augment the training data.…”

“…Recent examples include nucleophilic 2,3 and electrophilic aromatic substitution reactions, [4][5][6][7][8][9] Diels-Alder reactions, 10,11 Heck reactions, 12 radical C-H functionalisation of heterocycles, 13 and reactions such as alkylations, Michael additions, and aldol condensations that proceed through proton abstraction. 14 These methods have been based on either quantum chemical (QM) calculations, 2,5,6 machine learning (ML) trained on experimental data, 8,[10][11][12] or a combination of the two where QM has either provided descriptors for the ML model 3,9 or was used to augment the training data. 13,14 However, these approaches have rarely been compared on the same dataset.…”

Machine Learning Prediction of Structure‐Performance Relationship in Organic Synthesis

“…[ 39 ] Grzybowski developed an innovative approach which utilized the transition state‐like geometries and realized quantified stereoselectivity prediction in Diels‐Alder reactions and Michael addition reactions. [ 40 ] Jensen et al . established an intriguing on‐the‐fly strategy to achieve end‐to‐end selectivity prediction.…”

“…Benefiting from this merit, machine learning applications in chemistry have lately seen significant progress in a wide array of areas. [ 36‐60 ] For reaction performance prediction, Sigman's multi‐ variant linear regression approach has gained wide popularity and reached remarkable success in a series of organic transformations. [ 16,36‐37 ] Doyle's random forest model achieved excellent reaction yield prediction in Buchwald‐Hartwig cross coupling reactions.…”