Silvia Lanteri

A chemometric class modeling strategy (unequal dispersed classes [UNEQ]) is applied for the first time to evaluate harmful alcohol drinking within large population screening programs, in comparison with traditional strategies of data interpretation. Five inexpensive indirect biomarkers (aspartate aminotransferase, alanine aminotransferase, γ-glutamyltransferase, mean corpuscular volume and carbohydrate-deficient transferrin) were determined in blood samples from 423 patients, classified as low-risk or harmful drinkers, according to their ethanol consumption. The multivariate UNEQ approach remarkably improves the diagnostic performances of indirect biomarkers in harmful drinking evaluation, leading to reliable decision rules, with few doubtful classifications to be reviewed through complex confirmation procedures. This UNEQ model represents an innovative general approach for clinical evaluation that efficiently extracts the information content present in each biomarker to provide a new synthetic multivariate parameter, to be directly used in diagnostic protocols.

The present research is aimed at introducing a suitable approach for the exploitation of the hyperspectral data obtained by μATR-FTIR analyses of paint cross-sections. The application of principal component analysis for chemical mapping is well-established, even if a very limited number of applications to μFTIR data have been reported so far in the field of analytical chemistry for cultural heritage. Moreover, in many cases, chemometric tools are under-utilized and the outcomes under-interpreted. As a consequence, results and conclusions may be considerably compromised. In an attempt to overcome such drawbacks, the present work is proposing a comprehensive and efficient procedure based on an interactive brushing approach, which combines the structural information of the score scatter plots and the spatial information of the principal component (PC) score maps. In particular, the study demonstrates not only how the multivariate approach may provide more information than the univariate one, but also how the integration of different chemometric tools may allow a more comprehensive interpretation of the results with respect to the studies up to now reported in the literature. The examination of the average spectral profile of each score cluster, jointly with the loading analysis, is functional to characterize each area investigated on the basis of its spectral features. A multivariate comparison with spectra of standard compounds, projected in the PC score space, helps in supporting the chemical identification. The approach was validated on two real case studies.