Expert System for Predicting Reaction Conditions: The Michael Reaction Case

Abstract

A generic chemical transformation may often be achieved under various synthetic conditions. However, for any specific reagents, only one or a few among the reported synthetic protocols may be successful. For example, Michael β-addition reactions may proceed under different choices of solvent (e.g., hydrophobic, aprotic polar, protic) and catalyst (e.g., Brønsted acid, Lewis acid, Lewis base, etc.). Chemoinformatics methods could be efficiently used to establish a relationship between the reagent structures and the required reaction conditions, which would allow synthetic chemists to waste less time and resources in trying out various protocols in search for the appropriate one. In order to address this problem, a number of 2-classes classification models have been built on a set of 198 Michael reactions retrieved from literature. Trained models discriminate between processes that are compatible and respectively processes not feasible under a specific reaction condition option (feasible or not with a Lewis acid catalyst, feasible or not in hydrophobic solvent, etc.). Eight distinct models were built to decide the compatibility of a Michael addition process with each considered reaction condition option, while a ninth model was aimed to predict whether the assumed Michael addition is feasible at all. Different machine-learning methods (Support Vector Machine, Naive Bayes, and Random Forest) in combination with different types of descriptors (ISIDA fragments issued from Condensed Graphs of Reactions, MOLMAP, Electronic Effect Descriptors, and Chemistry Development Kit computed descriptors) have been used. Models have good predictive performance in 3-fold cross-validation done three times: balanced accuracy varies from 0.7 to 1. Developed models are available for the users at http://infochim.u-strasbg.fr/webserv/VSEngine.html . Eventually, these were challenged to predict feasibility conditions for ∼50 novel Michael reactions from the eNovalys database (originally from patent literature).

Modelling Chemical Reasoning to Predict and Invent Reactions

“…While the positive evaluation of a reaction prediction system can be easily achieved with a test set of hold‐out known reactions (as above), negative evaluation with reactions that are known not to occur is a difficult task because current publishing customs in organic chemistry do not provide incentives for publishing failed reactions or the limitations of synthetic methodology. This lack of data has been criticised both by synthetic chemists and chemoinformaticians …”

“…Systems that predict reaction conditions are seldom reported and are typically designed to predict specific reaction classes, for example, the Michael reaction …”

“…[13,14] HTRP with purely quantum-chemical or multi-scale approaches is yet to be conducted due to the tremendousc omputational cost of these methods. High-throughputr eactionp redictions are therefore performed with systems that use am odel of chemical knowledge in some form or another.T hey have been conductedw ith rule-based expert systems, machine-learning, formal logic and combinationsthereof with force fields or semiempirical methods: [6,[14][15][16][17][18][19][20][21][22][23][24][25][26][27][28] Expert systems based on manually enteredo ra lgorithmically extracted reaction rules are the mostw idely used approacht o predict reactions. [14][15][16][17] They are appealing because they are interpretable and resemble the way undergraduateo rganic chemistry is taught.H owever, they suffer from several disadvantages:1 )they require knowledge to be manuallye ncoded by experts either directly as reactionr ules or indirectly as complex heuristics to extract rules from data, 2) they becomec omplicated to maintain with ag rowingk nowledge base, and 3) they cannotgeneralise.…”

“…[16,29] HTRP with machinel earning has recently regained traction with severali nteresting approaches by the groups of von Lilienfeld, [21,22] Aires-de Sousa, [18,19] Aspuru-Guzik [23] and Varnek. [20] However,a lthough these systemsp erform well within their intended domaino fa pplicability,t hey are either limitedi ns cope or have not yet been shown to generalise to completely novel reactiontypes, especially with regards to transition-metal catalysis. In addition, they are not easily interpretable.…”

“…Systems that predictr eaction conditions are seldom reported and are typically designedt op redict specific reaction classes, forexample, the Michael reaction. [20] Baldi and co-workers reported an eleganta pproacht om ake predictions at the mechanistic level for reactions not involving transition-metalc atalysis. [16,17] However,e ven though mechanisms undoubtedly play an important role in understanding chemicalr eactions, many reactions can be predicted just by their overall transformation.D etailed mechanismsf or many reactions are not known and reliable data is often unavailable.…”