An Expert System for the Auto Assignment of 1H NMR Spectra of Small Molecules

POSTER by
1Carlos Cobas, 1Michael Bernstein, 1Esther Vaz, 1Felipe Seoane, 1Maruxa Sordo,
1Santiago Domínguez, 1Manuel Pérez, and 2Stanislav Sykora

1Mestrelab Research, Santiago de Compostela, Spain
2Extra Byte, Castano Primo, Italy

Presented at the
53rd Experimental NMR Conference (ENC), April 15-20, 2012, Miami, Florida (USA).

DOWNLOAD full poster: PDF (676 kB) Other NMR Articles and Posters | Stan's HUB

Abstract

In this work we present an integrated new software solution aimed at the automatic assignment of 1H NMR spectra of small molecules. It is based on an expert system constructed on principles of fuzzy logic with probabilistic methods used for the classification of the different resonances in the spectrum as well as for the enumeration of all the most likely assignments in the spectrum.

This expert system includes a number of sophisticated algorithms ranging from GSD [1] for the automatic deconvolution of peaks in the spectrum, to a fuzzy-logic based peaks classifier [2] that identifies NMR signals according to several predefined classes (compound, solvent, impurities, 13C satellites, amongst others).

Once a high quality list of classified peaks is available, all potential assignments are enumerated in such a way that each assignment is scored using a novel scoring system that takes into account not only the quality of the predicted chemical shifts and integrals compared to the experimental values, but also other spectral descriptors such as the kurtosis and root mean square width of every multiplet in the spectrum (both predicted and experimental). Finally, inter-multiplets connectivity are determined using the J-Correlator algorithm [3].

In this poster we shall illustrate different applications of this expert system. The results obtained when the algorithm was applied to a large number of typical organic molecules of medium complexity and molecular mass in a fully unattended manner show that more than 75% of all protons in the test set are successfully and correctly assigned. Some of the representative failures detected will also be illustrated.

Finally, the expert system has been designed to take into account the possibility of incorporating additional NMR experiments (e.g. 2D correlation spectra).

The results obtained when the algorithm was applied to a large number of typical organic molecules of medium complexity and molecular mass show that more than 75% of all protons in the test set are successfully and correctly assigned.


Please, cite this online document as:
Cobas C., Bernstein M., Vaz E., Seoane F., Sordo M., Domínguez S., Pérez M., Sykora S.,
   An Expert System for the Auto Assignment of 1H NMR Spectra of Small Molecules,
   Poster at 53rd ENC, Miami, April 15-20, 2012, DOI: 10.3247/SL4Nmr12.001.


References

[1] Carlos Cobas, Stanislav Sykora, The Bumpy Road towards Automatic Global Spectral Deconvolution (GSD), 50th ENC Conference, Asilomar, CA (USA), March 29-April 4, 2009. DOI: 10.3247/SL3Nmr09.003
[2] Cobas C., Seoane F., Dominguez S., Sykora S., A new approach to ... Global Spectral Deconvolution (GSD),
     Spectroscopy Europe, 23(1), 25-30, (2011).
[3] Carlos Cobas, Pablo Monje, Santiago Fraga, Stanislav Sykora, Novel Data Evaluation Algorithms: J-Correlator, 49th ENC Conference, Asilomar, CA (USA), March 9-14, 2008. DOI: 10.3247/SL2Nmr08.007


Discussions

Your comments are welcome and will appear here

Sep 1, 2013:
The present statistical performance on very real spectra is better than 86% correct assignments.

 

Stan Sykora's Publications, Posters and Courses & Talks Stan's Library | Stan's HUB
Copyright ©2012 Sýkora S.,    DOI: 10.3247/SL4Nmr12.001 Page design by Stan Sýkora