Science is complicated. The huge amount of data that needs to be recorded, analysed and understood has led to a push for better systems to automate and improve these processes. One such process is chemical analysis through nuclear magnetic resonance (NMR), and this is where “Ask Ernö,” and its creators from Ecole Polytechnique Fédérale de Lausanne step in.
As described in their paper, published this month in SpringerOpen’s Journal of Cheminformatics, automating NMR analysis is tricky. Firstly you have to get a system to be able to predict what an NMR spectrum of a given molecule will look like. Secondly, the system will have to be able to elucidate a molecular structure from a given spectra. Adding all this data to a program or system would take far too long to be practical, so Luc Patiny and colleagues designed a system, named “Ask Ernö” after Ernö Pretsch (who literally wrote the book on the necessary information for humans to assign spectra) that can not only do the above, but automatically improves itself through ‘machine learning.’
It must be emphasized that Ask Ernö developed this capability fully autonomously: at no point it was fed with the fruits of the labor of human experts. The learning process of Ask Ernö is akin to that of a newcomer to the realm of NMR analysis, who is told the basic rules of assignment and through experience and induction develops his own NMR tables.”
Castillo et al.
As explained by the authors, Ask Ernö works by automatically assigning a nucleus in a molecule to associate a substructure (the nucleus and its surroundings) with an observed chemical shift. As this information is stored in a database, it is used to predict further chemical shifts. As the initial database grows, the accuracy of these predictions improves. These in turn provide better chemical shift constraints to be used in subsequent assignments.
Ask Ernö can therefore ‘learn’ by running repeated assignment cycles on a given training set, using each new assignment to improve its predictions in the next cycle. As the authors report, after just 10 cycles, Ask Ernö was able to decrease its prediction error by 17%.
This whole process is fully autonomous: while the research was done on a ‘training set’ of over 2000 molecules, the researchers hope to develop Ask Ernö “into a state-of-the-art tool for automatic NMR analysis in the near future.”
You can read the full article, “Ask Ernö:” a self-learning tool for assignment and prediction of nuclear magnetic resonance spectra” on the Journal of Cheminformatics website.