Quantifying pairwise similarity of ignitable liquids analyzed by GC-MS: A comparison of four methods

For analyses that require the extraction and concentration of ignitable liquids, quantitatively comparing the similarity between samples obtained under experimental conditions to reference samples provides a straightforward way of defining the experimental response. Various methods for pairwise comparison of gas chromatography-mass spectrometry (GC-MS) data have been applied in the fire debris literature that operate on different data structures and employ different underlying metrics. In this article, we consider four of these methods (covariance mapping, summed ion similarity, Pearson product-moment correlation [PPMC], cosine similarity) and assess their performance using a dataset of replicate gasoline samples with known relationships. Method performance is considered separately for both selected ion monitoring data and full scan data and is evaluated by examining how well the distributions of various comparison types adhere to the known sample relationships. Method performance is found to be generally consistent across the two data types, but performance differs across methods. It is shown that all methods are invariant to data normalization, but the performance of the PPMC and cosine similarity methods is improved with a fourth-root data transformation. For the data considered here, the method of covariance mapping most closely conformed to expected performance based on the known sample relationships.