Analysis of complex mixtures by gas chromatography/mass spectrometry using a pattern recognition method

Abstract

the volatile trimethyl borate by codistillation with methanol, and the residue was heated 2 h at 85 "C, in vacuo to completely convert the aldonic acids into aldonolactones. Lower temperatures and/or shorter time periods resulted in partial conversion and the isolation of some free aldonic acids. Complete conversion is necessary because aldonic acids will not form N-propylaldonamide when treated with l-propylamine. Even though the amine is very basic, it did not cause epimerization under the conditions studied. Acetylation with pyridine and acetic anhydride requires 1 h and is quantitative. Dehydration reactions occur with the lactone (16) but not with the amide.Essentially the original aldoses have been converted into alditol acetates, and the original aldonic acids and aldonolactones have been converted into corresponding N-propylaldonamides. Calibration curves constructed by plotting detector response vs. milligrams of sample (1.2-6.7 mg) have a correlation coefficient greater than 0.995. A wider range of sample sizes can be accommodated by suitable concentration or dilution of the injected sample.There are a number of advantages to the new procedure. The method is able to handle complex mixtures of aldonic acids and aldoses. Alduronic acids can be determined by suitable modifications. Because the carboxyl group is kept, the stereochemical configuration of the sugar is maintained, unlike the differential method (2) wherein the lactone is reduced to the alditol. The new method adds a degree of versatility not found in most methods. The N-propyl group can be substituted easily by an N-butyl, N-amyl, or even an N-hexyl by replacing 1-propylamine by some other amine. The lactone will react under basically the same conditions to give a homologous derivative. The advantages are readily apparent in that the larger the N-alkyl group, the greater the elution time relative to the alditols or other possible compo-nents, e.g., phenyl P-D-ghcopyranoside. In fact, N-propylxylonamide coelutes with phenyl 0-D-glucopyranosides; by changing to N-amylamine, one is able to separate them into two peaks while maintaining the resolution seen in Figure 2. (5) Dmitriev, B. A.; Backinowsky, L. V.; Chizhov, 0. S.; Zoiotarev, 8. M.; Kochetkov, N. K. Analysis of a complex mlxture Involves a gas chromatographlc separation followed by a mass spectral search for Identification of each component. Thls Is an extremely tedlous andtime-consumlng process for fragrances and flavors whlch sometimes contaln hundreds of components. Thls presentatlon describes a computer pattern recognttlon method for the rapid ldentlficatlon of components in a complex mixture, the comparlson between the complete GC profiles of different samples, and the recognltlon of a group of compounds as a single entity. The central part of thls method is a comparlson algorlthm which compares the components In two mixtures uslng the K nearest neighbor classification rule. I t is Illustrated by comparing essential oils of dlfferent sources and by detectlng essentlal 011s In perfum...

Unraveling ingredients in complex mixtures by chromatographic spectrum recognition: Application to perfume deformulation

“…To overcome the limit due to the instability of retention indices to construct the molecular profiles of Chien, we constructed chromatographic spectra, that is, molecular profiles with a specificity axis ( x ), associating a different number to each molecular entity corresponding to the occurrence rank of the molecule in the database. Therefore, the molecular entities near the x ‐axis origin are the least specific because they are common to many ingredients of the database and those on the right are the most specific because they are present in only a few ingredients of the database.…”

“…To appraise the usefulness of chromatographic spectra to differentiate the ingredients in the ESO 2000 database, the similarity of the chromatographic spectra between and within classes was studied, by calculating similarity indices according to Equation (1). The similarity index, initially proposed by Chien, represents the probability that the chromatographic spectrum of ingredient 1 is contained in the chromatographic spectrum of ingredient 2. Within‐class similarity was studied by comparing the chromatographic spectra of the ingredients in each class and the chromatographic spectrum of their class by calculating similarity indices as described above.…”

“…The similarity between two chromatographic spectra was evaluated by calculating a similarity index proposed by Chien based on the classification method K nearest neighbors : Equation (1). The closer this index is to 1000, the more recognizable the chromatographic spectrum of ingredient 1 is in the chromatographic spectrum of ingredient 2.with: = concentration of molecular entity i of ingredient 1, = concentration of molecular entity i of ingredient 2, = greatest possible distance between ingredient 1 and ingredient 2, = number of molecular entities common to ingredients 1 and 2, = number of molecular entities in ingredient 1, and = weighting factor.…”

“…Chien proposed an alternative approach based on recognizing a set of molecular entities (not necessarily distinctive) forming the signature of an ingredient. The aim was to automate the deformulation process and so facilitate and speed up this lengthy manual task.…”

“…Unknown mixtures are thus compared with ingredients only for the 10‐15 main molecular entities on the chromatogram, which again speeds up the process. The approach proposed by Chien allowed four ingredients in a six‐ingredient mixture to be recognized. Tomàs et al adopted a similar principle, targeting 12 molecular entities to deformulate mixtures using the simplex method to solve the model's linear combination.…”