Data Enhanced Reaction Predictions in Chemical Space With Hammett's Equation
By separating the effect of substituents from chemical process variables, such as reaction mechanism, solvent, or temperature, the Hammett equation enables control of chemical reactivity throughout chemical space. We used global regression to optimize Hammett parameters $ρ$ and $σ$ in two datasets, experimental rate constants for benzylbromides reacting with thiols and the decomposition of ammonium salts, and a synthetic dataset consisting of computational activation energies of $\sim$ 1400 $S_N2$ reactions, with various nucleophiles and leaving groups (-H, -F, -Cl, -Br) and functional groups (-H, -NO$_2$, -CN, -NH$_3$, -CH$_3$). The original approach is generalized to predict potential energies of activation in non aromatic molecular scaffolds with multiple substituents. Individual substituents contribute additively to molecular $σ$ with a unique regression term, which quantifies the inductive effect. Moreover, the position dependence of the substituent can be replaced by a distance decaying factor for $S_N2$. Use of the Hammett equation as a base-line model for $Δ$-Machine learning models of the activation energy in chemical space results in substantially improved learning curves for small training set sizes.