Predicting structural groups of small molecules from 1H NMR spectral features using common machine learning classifiers

Structural determination of molecules using solution-state nuclear magnetic resonance (NMR) is a time-consuming effort mostly due to spectral analysis and correlation of spectral features with structural motifs. A few machine learning methods exist to aid this step of the workflow, requiring at least 1H and 13C chemical shifts to make predictions. In this paper we show that it is possible to predict, with good accuracy (> 0.8), structural groups of small molecules using only 1H NMR spectral features (chemical shift, J-coupling, integral values, and splitting patterns). For this task we employed common machine learning classifiers found in the sklearn python module, and a database constructed using 1H NMR spectra found in online NMR tools (NMR-Challenge and NMRium).