Assessing the effect of data pretreatment procedures for principal components analysis of chromatographic data

Abstract

Following publication of the National Academy of Sciences report “Strengthening Forensic Science in the United States: A Path Forward”, there has been increasing interest in the application of multivariate statistical procedures for the evaluation of forensic evidence. However, prior to statistical analysis, variance from sources other than the sample must be minimized through application of data pretreatment procedures. This is necessary to ensure that subsequent statistical analysis of the data provides meaningful results.The purpose of this work was to evaluate the effect of pretreatment procedures on multivariate statistical analysis of chromatographic data obtained for a reference set of diesel fuels. Diesel was selected due to its chemical complexity and forensic relevance, both for fire debris and environmental forensic applications. Principal components analysis (PCA) was applied to the untreated chromatograms to assess association of replicates and discrimination among the different diesel samples. The chromatograms were then pretreated by sequentially applying the following procedures: background correction, smoothing, retention-time alignment, and normalization. The effect of each procedure on association and discrimination was evaluated based on the association of replicates in the PCA scores plot. For these data, background correction and smoothing offered minimal improvement, whereas alignment and normalization offered the greatest improvement in the association of replicates and discrimination among highly similar samples. Further, prior to pretreatment, the first principal component accounted for only non-sample sources of variance. Following pretreatment, these sources were minimized and the first principal component accounted for significant chemical differences among the diesel samples. These results highlight the need for pretreatment procedures and provide a metric to assess the effect of pretreatment on subsequent multivariate statistical analysis of complex data.