ACD/Labs 5^th Annual European Users' Meeting 2004 (EUM)

Strategies on Building Training Databases for Optimizing NMR Prediction with ACD/Labs Software: The First Step toward Reliable Automatic Assignment

Stelian Nicula, Daniela Borghi

Abstract

The aim of this work is to provide a reliable tool in hands of occasionally NMR users as well as in hands of NMR spectroscopists working in analytical labs. Much progress has been made in recent years in automation of NMR spectral analysis and interpretation. Precise prediction is a sine qua non condition for reliable automatic analysis of spectra. Several algorithms were designed in order to handle various workflows and the need of a manual intervention against a complete automatic analysis of NMR spectra depends on many individual factors. Our laboratory provides a complete NMR service to the chemists by analyzing spectra of more that 5000 samples/year. An SQL database of accurately assigned NMR spectra is provided containing useful NMR information for further analysis of compounds when elaborating a structural class. The ACD/Labs' predicting capabilities gives important support to our chemists recording NMR spectra in open access and stimulated the definition a strategy for optimizing the spectra prediction. One of the first question to be answered was how large, in terms of structures, must be a user database of NMR frequencies in order to assure a consistent improvement of the prediction? Is the amount of information available predominant with respect to the quality? We tried to answer to these questions by exploiting the large amount of NMR data available on our database of assigned spectra. The prediction improvement against the number of entries in the user db is depicted in the figure.

Download the presentation in MS PowerPoint (2.69 Mb ZIP file) or Adobe Acrobat format (699 Kb PDF file).

Relevant Products/Product Lines: NMR