Discovering predictive rules of chemistry from property landscapes

Predicting the chemical and physical properties of molecules often relies on systematic rules relating the properties to molecular characteristics. This Letter introduces a novel method to reveal predictive chemical rules based on analysis of the chemical property landscape, which specifies the functional relationship between a measured property and an appropriate set of molecular variables. As an illustration, we consider landscapes relating the 17O NMR chemical shift, 13C NMR chemical shift, and IR vibrational frequency to the moieties attached to a carbonyl group. Implications of this ‘Chemscape’ formulation for general molecular property prediction are discussed.