Predicting chance infrared spectroscopic matching frequencies Original Research Article

Identification by spectroscopic matching is important in commercial and forensic analysis. The occurrence of chance matches of target infrared spectra is investigated by intercomparison within a spectral database of approximately 50 000 materials. While the frequency of chance matches varies substantially with the material of interest, a simple binomial model gave predictions up to seven orders of magnitude too low for a 6-peak match, while hypergeometric probabilities improved the prediction to within one order of magnitude of observation. It was necessary to take account of both the effective number of possible peak positions, and the distribution of number of peaks within spectra, to obtain reasonable predictions. Simple truncation of the number of positions, coupled with assumed distributions of number of peaks based on observed parameters for a small sample from the database, gave predictions within a factor of two of observation; the predictions were essentially exact given a large sample estimate of the mean peak incidence probability. The implications for confirmatory analysis are discussed with particular reference to the detection of veterinary drug residues.