Effects of data pretreatment on the multivariate statistical analysis of chemically complex samples.

详细信息   

• 作者：McIlroy ; John William.
• 学历：M.S.
• 年：2014
• 毕业院校：Michigan State University
• Department：Forensic Science
• ISBN：9781303831454
• CBH：1554052
• Country：USA
• 语种：English
• FileSize：14230224
• Pages：189

文摘

Multivariate statistical procedures,such as principal component analysis PCA),are often utilized to differentiate and associate a large number of complex samples consisting of thousands of variables. When samples with similar chemical compositions are compared,chemical differences between samples are often overshadowed by non-chemical variation. Therefore,in order to provide meaningful statistical comparisons and differentiate complex and highly similar samples,these non-chemical sources of variation must be minimized,often accomplished by implementing data pretreatment procedures. In this work,ten diesel samples from different service stations were analyzed in triplicate by gas chromatography-mass spectrometry. The resulting chromatograms were processed with data pretreatment procedures,including baseline correction,smoothing,retention time alignment,and normalization,to evaluate the enhanced discrimination in PCA achieved by minimizing non-chemical variation. For each pretreatment procedure,metrics were developed to evaluate the effect on the chromatogram as well as the PCA results. Normalization and alignment resulted in the greatest enhancement in association of replicate samples,while smoothing and baseline correction were shown to have minimal effect. By applying data pretreatment procedures,replicate samples were closely associated with one another and differentiated from the other diesel samples,allowing for differentiation of complex and similar samples.