par Praisler, Mirela;Dirinck, I.;Van Bocxlaer, Jan F P J.;De Leenheer, André;Massart, Désiré Luc 
Référence Talanta, 53, 1, page (155-170)
Publication Publié, 2000-10
Référence Talanta, 53, 1, page (155-170)
Publication Publié, 2000-10
Article révisé par les pairs
| Résumé : | A computer-aided procedure automating the identification of illicit amphetamine analogs eluting from a gas chromatograph coupled to a Fourier transform infrared spectrometer is presented. The expert system discriminates novel amphetamines from other classes of drugs of abuse normally screened in illicit tablets or powders. The main analytical advantages of the system over the automated procedures dedicated to general unknown analysis are the objectivity and the accuracy in predicting the class identity of the compound (i.e. stimulant, hallucinogen) when the reference spectrum is not present in the spectral library. The expert system uses quantitative thresholds defining the similarity of the unknown to the classes of illicit amphetamines and checks the presence of the molecular skeletons associated with different psychotropic effects of amphetamines. The challenge in building the system was the fuzziness of vapor-phase Fourier transform infrared spectrometer spectra of low-weight molecules such as amphetamines. This paper emphasizes the chemometrical techniques found most appropriate for modeling such spectral behavior. An exploratory (principal component) analysis indicated the sample preparation and the feature weight function yielding the best input for the knowledge base. The class identity of a compound was assigned using Soft Independent Modeling of Class Analogy. A rule-based decision system was implemented to enhance the accuracy in identity assignment. The flow diagram optimizing the knowledge base content of each model is presented. Finally, up to 81.13% (out of 159 tested compounds) were classified with a 5% confidence level. The total correct classification rate was 93.93%, for a yield of 96.30% true positive amphetamines. A computer-aided procedure automating the identification of illicit amphetamine analogs eluting from a gas chromatograph coupled to a Fourier transform infrared spectrometer is presented. The expert system discriminates novel amphetamines from other classes of drugs of abuse normally screened in illicit tablets or powders. The main analytical advantages of the system over the automated procedures dedicated to general unknown analysis are the objectivity and the accuracy in predicting the class identity of the compound (i.e. stimulant, hallucinogen) when the reference spectrum is not present in the spectral library. The expert system uses quantitative thresholds defining the similarity of the unknown to the classes of illicit amphetamines and checks the presence of the molecular skeletons associated with different psychotropic effects of amphetamines. The challenge in building the system was the fuzziness of vapor-phase Fourier transform infrared spectrometer spectra of low-weight molecules such as amphetamines. This paper emphasizes the chemometrical techniques found most appropriate for modeling such spectral behavior. An exploratory (principal component) analysis indicated the sample preparation and the feature weight function yielding the best input for the knowledge base. The class identity of a compound was assigned using Soft Independent Modeling of Class Analogy. A rule-based decision system was implemented to enhance the accuracy in identity assignment. The flow diagram optimizing the knowledge base content of each model is presented. Finally, up to 81.13% (out of 159 tested compounds) were classified with a 5% confidence level. The total correct classification rate was 93.93%, for a yield of 96.30% true positive amphetamines. © 2000 Elsevier Science B.V. |
