Uninformative Variable Elimination Combined With Three Spectroscopy Methods For Quantitive Analysis

Quantitative analysis by near-infrared (NIR) spectroscopy, ultraviolet (UV) spectroscopy and mass spectroscopy (MS) combined with partial least squares (PLS) were studied. In order to get an optimized modeling, uninformative variable elimination wae used to screening uninformative variables and eliminate noise and interference information.For UV spectroscopy, p-cyclodextrin was used as chiral reagents to measure the ratio of phenylalanine and metalaxyl isomers.To overcome the effects of the total concentration of phenylalanine and metalaxyl on the model, we took measures of concentration normalization and reference band normalization to correcte the two original spectra.Quantitative analysis models of the two isomers were established. After UVE selecting the variables, two quantitative models of phenylalanine-β-cyclodextrin and metalaxyl-β-cyclodextrin were developed and optimized.Since the UV absorption has good linear correlation with the concentration of the sample, there are too many variables chosen by UVE, which contains a large number of linear correlation variables. Successive projections algorithm (SPA) has a good ability to reject the linear correlation variables, so SPA was used for further screening and a more simplier model of phenylalanine and metalaxyl isomer were established.For IR spectroscopy,40liquid samples were determined combined with partial least squares method.2,2,4-trimethylpentane,n-heptane, cyclohexane, ethyl acetate and butyl acetate were determined respectively by the full spectrum with PLS modeling and UVE screening spectrum with PLS modeling. The163solid samples of tabacco powder were also determined combined with partial least squares method. Total nitrogen and total suger in tobacco powder were determined respectively by the full spectrum with PLS modeling and UVE screening spectrum with PLS modeling.For MS, to measure the concent of2,2,4-trimethyl pentane, n-heptane, butyl acetate, toluene and acetophenone with tatal40samples by MS spectrometry combined with partial least squares. Firstly, the spectral obtained by GC-MS were being accumulated on the time direction to get simulated mass spectral data of the mixture.Secondly, establish a quantitative analysis of the five compounds by two methods,one is choosing specified variables of m/z to establishing the model,the other is establishing the model with the variables screening after UVE. The research and results described as follows:(1) β-cyclodextrin was used as chiral reagents to determining the molar fraction of27phenylalanine samples and60metalaxyl samples optical isomer with PLS. After concentration normalization of phenylalanine:84.92%variable were slected by UVE algorithm and established a quantitative model, root mean square error of prediction set(RMSEP) was0.039;2.15%variable were selected by UVE-SPA modeling, RMSEP was0.013; After reference normalization of phenylalanine:70.77%variable were slected by UVE algorithm and established a quantitative model,RMSEP was0.022;1.23%variable were selected by UVE-SPA modeling, RMSEP was0.0070; After concentration normalization of metalaxyl:33.18%variable were slected by UVE algorithm and established a quantitative model, RMSEP was0.018;1.23%variable were selected by UVE-SPA modeling, RMSEP was0.022; After reference normalization of metalaxyl:73.43%variable were slected by UVE algorithm and established a quantitative model, RMSEP was0.019;1.38%variable were selected by UVE-SPA modeling, RMSEP was0.0090.Therefore, it is useful of UVE-SPA optimize and simplify the UV-visible spectroscopy by modeling.(2)For the liqid samples:five compositon of2,2,4-trimethylpentane, n-heptane, cyclohexane, ethyl acetate and butyl acetate,RMSEP were0.029,0.18,0.038,0.010,0.013after UVE. For solid samples:tobacco powder:RMSEP of total nitrogen was reduced from0.18to0.11after UVE; total suger of RMSEP was reduced from2.2to0.78after UVE.Near-infrared quantitative model results show that UVE model optimization effect is obviously.(3) Analysis the composition of2,2,4-trimethyl pentane, n-heptane, n-butyl acetate, toluene and acetophenone. m/z specified variables selection:RMSEP were respectively0.048,0.11,0.021,0.030,0.009.After UVE screening:RMSEP were respectively0.035,0.12,0.014,0.042,0.0090. The results showed that, the mixture MS spectra were possible for quantatively analysis. UVE method for optimizing the model is obviously.Quantitative analysis of the mixture using UV, NIR and MS, the optimization ariables selected have achieved significant results. UVE variable selection played an important role in modeling.Therefore, this method will have broadly application prospects in the field of quantitative analysis of the mixture.