Novel Chemometric Models And Methods For Quantitative Spectroscopic Analysis Of Complex Systems

This thesis aims at developing novel chemometirc models and methods forquantitative spectroscopic analysis of complex systems. It mainly involves thefollowing aspects:When analyzing complex heterogeneous mixtures that exhibit sample-to-samplevariability in physical properties using spectroscopic instrumentation, the uncontrolledphysical differences between samples (e.g., particle size and shape, sample packing,and sample surface, etc.) usually cause multiplicative light scattering effects on themeasured spectrum. The separation of the spectral contributions due to variations inchemical compositions from those caused by multiplicative effects is therefore crucialfor the accurate quantitative analysis of messy spectral data with multiplicative effects.In Chapter2, a modified version of the optical path length estimation and correction(OPLECm) method was proposed to correct the multiplicative effects in spectralmeasurements of heterogeneous system. The performance of the proposed method wastested on the near-infrared spectral data of two systems (i.e., near-infrared total diffusetransmittance spectra of four-component suspension samples and near-infrared spectraldata of meat samples). The experimental results showed that OPLECmcombined withthe Dual Calibration Strategy (DCS) realized accurate quantitative spectroscopicanalysis for the analytes of interest in heterogeneous mixture samples even in thepresence of multiplicative light scattering effects.NIR spectroscopic technique coupled with dry film method offers a rapid,reagent-free and non-destructive alternative for the determination of plasma glucoselevels. However, the uncontrollable variations in the thickness and light scatteringcharacteristics of dry film samples introduce significantly multiplicative spectralvariations on NIR transmission measurements of dried plasma samples, and hencehinder the practical application of the dry film-based NIR technique for blood glucoseassay. In Chapter3, a quantitative model was proposed to explicitly model the effectsof the variations in thickness and light scattering characteristics of dry plasma films onNIR transmission measurements. And an advanced dual calibration strategy based onsupport vector regression (DCSSVR) was derived from the proposed quantitative modelto realize accurate determination of analytes (e.g. glucose) in dried plasmas samples.The experimental results show that DCSSVRcould accurately predict the glucose concentrations in plasma samples from the NIR transmission measurements of theirdry films without the use of internal standard.Variations in the physical properties of enhancing substrates (e.g., size, shape andaggregation of Ag or Au nanoparticle colloids), the intensity and alignment/focusing oflaser excitation source exert a significant influence on the SERS intensities of theanalyte of interest. The production of highly stable and reproducible enhancingsubstrates (especially the widely used silver or gold nanoparticle colloids) is stillchallenging, which greatly hampers the practical quantitative application of SERStechnique. In Chapter4, a multiplicative effects model for quantitative SERS assays(MEMSERS) was developed. Based on the MEMSERSmodel, two detection modes, i.e.,internal standard addition detection mode and internal standard tagging detection mode,were proposed with an attempt to realize accurate quantitative SERS assays for routineuse. The experimental results on two model systems demonstrated that the combinationof MEMSERSmodel with internal standard addition detection mode or internal standardtagging detection mode could effectively alleviate the detrimental multiplicativeeffects due to the heterogeneity of enhancing substrates (Ag nanoparticle colloids), andachieved quite accurate concentration predictions for the analyte of interest (e.g.,R6G).Because of the high toxicity of nitrite to human body, the search for simple, rapidand selective method for the determination of nitrite is of great significance for publichealth and safety. In Chapter5, the MEMSERSmodel was further explored and appliedto the determination of nitrite in drinking water. Nitrite was first transformed into azodye by the diazotization-coupling reaction in the presence of p-nitroaniline anddiphenylamine under acidic conditions. The reaction products were then mixed withsilver nanoparticle colloids to induce SERS effect. The concentrations of nitrite ion inwater samples were indirectly determined from the SERS measurements of azo dye byMEMSERSmodel. The experimental results showed that MEMSERSmodel achievedquite accurate concentration predictions for nitrite ions with an average relativepredictive error less than5.0%. For water samples free of interferences, the recoveriesof nitrite ion were about100%. But for boiled tap water samples, the recoveries ofnitrite were much lower than100%. A reasonable explanation might be that theimpurities presented in boiled tap water samples occupied the adsorption sites on thesurfaces of silver nanoparticles, and hence reduced the adsorption of azo dye ontosilver nanoparticles to give rise to SERS signals. The development of chemometric software requires chemometric professionaltalents are proficient in both programming skills and mathematics, which makes thedevelopment of chemometric software quite difficult. In Chapter6, the hybridprogramming technology of Visual Basic/Visual C++and MATLAB was explored witha view to reduce the difficulty of software development and shorten the softwaredevelopment cycle. The F-4500three-dimensional fluorescence analytical applicationand three-dimensional data analysis console application demonstrated that hybridprogramming technique using Visual Basic/Visual C++and MATLAB not onlyshortened the software development cycle, but also promoted the software quality. Thischapter also discussed how to write the C++library (i.e., the core algorithms module inchemometrics software) for chemometric algorithms. Three base classes (vector class,matrix class and three dimension array class) were developed. Based on these baseclasses, some chemometric algorithm classes were developed to provide the corealgorithm modules for the follow-up development of chemometrics software usingVisual C++.