Part I. Synthesis and Application of α-Per(inter)halocarbonyl Compounds Part II. Development of Simple Predictive Models for Organic Reactions with Small Dataset and Minimal Features

Abstract

Part I. Synthesis and Application of α-Per(inter)halocarbonyl Compounds: Geminal chlorofluorides are versatile synthetic precursors to various organofluorines with a tetrasubstituted carbon center. In this work, we reported the geminal chlorofluorination of 1,2-dicarbonyl compounds via the tandem deoxygenative electrophilic and nucleophilic halogenations. The rationally designed dealkylation-resistant phosphoramidite enabled the use of non-proton electrophiles and heteroatom-bearing 1,2-dicarbonyl compounds. As a result, α-keto esters (O), α-keto thioesters (S), α-keto N-acylindoles (N), and α-keto acylsilane (Si) were successfully transformed to doubly or triply hetero-functionalized tetrasubstituted carbon centers with excellent site-selectivity. Subsequently, upon derivatization of geminal chlorofluorides with azide, we discovered the remarkable accelerating effect of the geminal fluorine substituent that enables the facile rearrangement of geminal azidofluorides into imidoyl fluorides without the typically required aid of strong acid under mild reaction conditions. The role of geminal fluorine was elucidated by both experimental and computational investigations. This new reactivity led to a practical one-step tandem preparative method for rarely known bench-stable imidoyl fluorides from a wide range of structurally diverse geminal chlorofluorides. To expand beyond 1,2-dicarbonyl substrates, the electronically similar, synthetically and pharmaceutically valuable α-perfluoroketones were evaluated. However, the inherent challenges associated with the activation and discrimination of the C–F bonds typically lead to over-defluorination as well as functional group incompatibility. We addressed these problems by utilizing our group’s rationally designed organophosphorus reagent that promoted mild and selective manipulation of single C–F bond in trifluoromethyl and pentafluoroethyl ketones via an interrupted Perkow-type reaction, which allowed the replacement of fluorine with more labile and synthetically versatile congeners such as chlorine, bromine, and iodine. The resulting α-halo-perfluoro ketones have two reactive units with orthogonal properties that would be suitable for the subsequent structural diversification. DFT calculation identified the favorable P–F interaction as the crucial factor for both thermodynamic and kinetic viewpoints. Part II. Development of Simple Predictive Models for Organic Reactions with Small Dataset and Minimal Features: Machine learning (ML) is an emerging area in organic synthesis for the reaction design and prediction. In recent studies, the ML approach for reaction development using big data with many features provided the best reaction conditions for the optimal yields and stereoselectivities. Despite the high performance, the preparation of large datasets is often difficult, especially for non-specialists. In this study, simple ML models were developed by utilizing easily available and familiar 13C NMR chemical shifts of the reacting sites in the substrates for our geminal chlorofluorination. Upon training on small datasets (<150) with condensed features, the feed-forward neural network (FNN) model could predict the yields and site-selectivities with reasonable efficiency. Moreover, we observed a notaable improvement in performance upon removal of empirically less relevant features. Subsequently, our ML model was advanced through the utilization of an unusual tabular augmentation method for stereoselective geminal bromofluoroolefination by fitting the real datasets into sigmoid or logarithmic curves. With the augmented dataset, the prediction of reaction profiles with the FNN model was substantially improved. The linearly combined use of our augmentation technique and conditional tabular generative adversarial network (CTGAN) also enhanced the model even further.