| Title: | "Fisher ratio feature selection by manual peak area calculations on comprehensive two-dimensional gas chromatography data using standard mixtures with variable composition, storage, and interferences" |
| Author(s): | Oliva DJ; Cheung C; Perrault KA; |
| Address: | "Laboratory of Forensic and Bioanalytical Chemistry, Chaminade University of Honolulu, Honolulu, HI, 96815, USA. Laboratory of Forensic and Bioanalytical Chemistry, Chaminade University of Honolulu, Honolulu, HI, 96815, USA. Katelynn.perrault@chaminade.edu" |
| DOI: | 10.1007/s00216-022-04484-8 |
| ISSN/ISBN: | 1618-2650 (Electronic) 1618-2642 (Linking) |
| Abstract: | "Comprehensive two-dimensional gas chromatography (GCxGC) is becoming increasingly more common for non-targeted characterization of complex volatile mixtures. The information gained with higher peak capacity and sensitivity provides additional sample composition information when one-dimensional GC is not adequate. GCxGC generates complex multivariate data sets when using non-targeted analysis to discover analytes. Fisher ratio (FR) analysis is applied to discern class markers, limiting complex GCxGC profiles to the most discriminating compounds between classes. While many approaches for feature selection using FR analysis exist, FR can be calculated relatively easily directly on peak areas after any native software has performed peak detection. This study evaluated the success rates of manual FR calculation and comparison to a critical F-value for samples analyzed by GCxGC with defined concentration differences. Long-term storage of samples and other spiked interferences were also investigated to examine their impact on analyzing mixtures using this FR feature selection strategy. Success rates were generally high with mostly 90-100% success rates and some instances of percentages between 80 and 90%. There were rare cases of false positives present and a low occurrence of false negatives. When errors were made in the selection of a compound, it was typically due to chromatographic artifacts present in chromatograms and not from the FR approach itself. This work provides foundational experimental data on the use of manual FR calculations for feature selection from GCxGC data" |
| Keywords: | Chemometrics Feature selection GCxGC Multidimensional gas chromatography Non-targeted analysis Volatile organic compounds (VOCs); |
| Notes: | "PubMed-not-MEDLINEOliva, Danson J Cheung, Cynthia Perrault, Katelynn A eng 1752607/National Science Foundation/ Germany 2022/12/16 Anal Bioanal Chem. 2023 May; 415(13):2575-2585. doi: 10.1007/s00216-022-04484-8. Epub 2022 Dec 15" |
