Predicting Experimental Yields as an Index to Rank Synthesis Routes

Abstract

It is possible to create novel synthetic routes for compounds using synthesis route design systems (SRDS) such as TOSP and KOSP. We have been investigating an in silico screening protocol which makes it possible to reduce the number of SRDS experiments in developing new synthesis routes. However, there still remains a problem on how to rank synthesis routes for experiments. Experimental yields are considered to be one of the most important factors in determining which synthesis route is better. If we can predict experimental yields before starting experimental works, it should be very helpful to rank the synthesis routes. The present study describes an attempt towards predicting the trends of experimental yields for organic synthesis by fusing computational chemistry and cheminformatics. A PLS regression was used to correlate experimental yields with the calculated activation energies Ea(calc), together with experimental conditions such as the dielectric constants of the solvents, reaction times, and reaction temperature as explanatory variables. The method was applied to expect experimental yields of two types of reaction, Diels-Alder reaction and Curtius rearrangements.