Towards automatically verifying chemical structures: the powerful combination of ¹H NMR and IR spectroscopy

Abstract

Human interpretation of spectroscopic data remains key to confirming new structures; the quest for speed and resource-efficiency suggests automating structure verification. We report that the combination of NMR and easily accessible IR data greatly improves its performance. We introduce an algorithm to quantify the similarity between experimental and calculated IR spectra and apply this to distinguish between a test set of 43 molecules and 100 similar isomeric structures. We describe a method to combine IR and ¹H NMR results measuring performance as the structure classification characteristic area over curve (SCC-AOC). Combination of IR and ¹H NMR significantly outperforms either technique alone (SCC-AOC 0.025 for combined data compared to IR 0.053 and 1H NMR 0.101 and a large step towards the ideal SCC-AOC value of zero). It drives the correct classification rate of the 100 comparisons to 87% from ca. 80% for individual methods and brings reliable automation within grasp.