A machine learning approach for predicting the fluorination strength of electrophilic fluorinating reagents

The unusual properties of a wide range of organofluorine compounds have provided strong incentives to the scientific community for the development of this field. In parallel to the constantly growing number of organofluorine compounds, an unusually high number of electrophilic N-F fluorinating reagents have emerged as potential fluorinators to achieve fluorine substitution in a simple and efficient manner. Bench stability, crystalline nature and modular synthesis are some of the key characteristics that make them increasingly important in synthetic transformations. In this context, it is important to understand the reactive power of these N-F fluorinating reagents in a quantitative manner. Experimental and DFT investigations to obtain a quantitative understanding of the fluorination power of these reagents are resource intensive, laborious and expensive. Herein, we propose a machine learning approach for predicting the relative power of a wide range of N-F fluorinating reagents by utilizing a simple and fast SMILES-based molecular encoding approach. A neural network algorithm was employed on a novel dataset consisting of four molecular descriptors, two categorical descriptors and 260 data points and was successful in predicting the fluorine plus detachment values for N-F fluorinating reagents belonging to six different categories.