An automated library search system for 13C-n.m.r. spectra based on the reproducibility of chemical shifts

Abstract

SUMMARYA library search system for 13C-n.m.r. spectra, based on a statistical description of the reproducibility of chemical shifts, is presented. A similarity index in the form of a significance probability (P-value) is developed from a previously introduced general concept. The applied data base of some 6000 spectra originates from the NIH-EPA Chemical Information System (CIS). The reproducibility model and the retrieval system are developed on a CDC Cyber 175 computer, with PASCAL as the programming language. The performance of the system is evaluated by using recall/reliability and confusion/ recall plots. In a tentative comparison with the Clerc search method (included in the CIS package), the Utrecht *fC-n.m.r. reproducibility-based retrieval system shows a better identification performance. The system is adaptable for use on a microcomputer.In recent years, several systems have been proposed for computer-aided library search of 13C-n.m.r. spectra, aiming at the identification of organic compounds and based on various coding and comparison algorithms [l-13]. Collections of several thousands of spectra, from different sources (i.e., recorded on different instruments, under various experimental conditions) are usually applied as data bases. Such data bases, which are often commercially available, generally suffer from poor interlaboratory reproducibility of the spectral data involved. In a previous paper [14], a new similarity index for "straightforward" library search methods was introduced, primarily for application to this type of data base. The main object of straightforward search methods is to retrieve the reference data (if available) of the unknown compound. This contrasts with "interpretative" methods, which aim principally at retrieving data of compounds similar in structure to the unknown. The proposed index has the form of a significance probability and is developed from a statistical model of the reproducibility of the quantities used for the comparison of unknown and reference data. Defined in general terms, this index is applicable to different types of analytical data, provided that the variables used to characterize unknown and reference spectra (feature quantities) are continuous in nature. In 13C!-n.m.r. spectra, the chemical shift and the peak intensity are such variables, in contrast to multiplicities. When a large multisource data base such as the CIS collection is used, how-

Identification and substructure analysis of oligosaccharide chains derived from glycoproteins by computer retrieval of high‐resolution 1H‐NMR spectra

“…NEUACa (2-6)GALb (1)(2)(3)(4) The module FIND allows the entering of a search key (spectrum identifier, compositional formula or chemicai shirt intervals), which results in a list of reference compounds containing the search key. In Figure 12 …”

“…Recall-reliability plots for SIeomb using preselections ( -) and using no preselections (---) 95-1 -60, 1-30-1 a m ) , each indicating the presence of a particular structure element, one of the eight subfiles is selected for retrieval. (2) Reference spectra for which a SImismatch value less than 2% is calculated are not further processed. As can be seen from Figure 10 Figure 12.…”

Pattern Recognition Used to Investigate Multivariate Data in Analytical Chemistry

“…Methods for file searching (49), spectral interpretation (50), chemo--metrics approaches (51), and additivity of contributions (37, 38) have appeared recently. Empirical studies in this area are justified because theoretical methods have not yet fully elucidated the origin of chemical shifts.…”