A DFT/machine-learning hybrid method for the prediction of 3 JHCCH couplings.

A machine learning model for the prediction of vicinal proton-proton couplings has been developed based on a hybrid representation that includes geometrical and electronic parameters derived from natural bond orbital (NBO) analysis of low-level BLYP/STO-3G computations. The model can predict 3 JHH couplings with accuracy comparable or better than the well-known Altona equation, and it can provide sensible 3 JHH predictions in systems not well handled by the Altona equation such as epoxide or cyclopropane rings.