Universal J-coupling prediction

Abstract

A data driven approach for small molecule J-coupling prediction is presented. The method is targeted for use as part of an automatic spectrum analysis, therefore emphasizing prediction coverage, maintainability, and speed in the design. The database search involves encoding the coupling path atom types into hash codes, which are used to retrieve the matching coupling constant entries from the database. The fast hash dictionary search is followed by a k Nearest Neighbors regression to resolve the substituent and conformational dependencies, parametrized with atomic charges, torsion angles, and steric bulk.