Large scale evaluation of a pattern recognition/expert system for mass spectral molecular weight estimation

Abstract

A previously developed fast, personal-computer based method of estimating molecular weights of organic compounds from low resolution mass spectra has been thoroughly evaluated. The method is based on a rule-based pattern recognition/expert system approach which uses empirical linear corrections which are iteratively applied to two mass spectral features to yield estimates. This technique has been extensively evaluated with 400 spectra of volatile and nonvolatile compounds of pharmaceutical interest and with 31378 high quality NIST spectra of compounds of molecular weight 30–500. Subsets of the NIST spectra were evaluated including a 23 989 spectra volatile set. The overall median and average absolute deviations from the true molecular weights of the 400 spectra were 1.5 and 13 daltons. For the large NIST set and volatile subset the overall median and average absolute deviations were 1.8–2.0 and 13–17 daltons. In both sets of spectra the best results were obtained with the nonhalobenzene and unknown classes. Due to misclassification errors better results were obtained by bypassing the classifier. Median errors with spectra with the molecular ion present were ca. twenty times lower than those without the molecular ion. The present system can rapidly produce molecular weight estimates with median absolute errors of 2 (average 15) daltons.