Identification of geographical origins of soybean pastes using headspace gas chromatography-mass spectrometry by selecting sample-descriptive components with an Incremental Association Markov Blanket

• Published in: Food research international 2023-12, Vol.174 (Pt 1), p.113492-113492, Article 113492
• Main Authors: Jeong, Seongsoo, Kwon, Dokyung, Lim, Jina, Jang, Hanbyeol, Kim, Jeongkwon, Chung, Hoeil
• Format: Article
• Language: English
• Subjects: Authentication; Gas Chromatography-Mass Spectrometry - methods; Geographical origin; Geography; Glycine max; Headspace gas chromatography-mass spectrometry; Incremental Association Markov Blanket; Ketones; Odorants; Soybean paste
• ISSN: 0963-9969; 1873-7145
• DOI: 10.1016/j.foodres.2023.113492

[Display omitted] •HS (head space)-GC-MS was adopted to identify the geographical origins of soybean pastes.•Thirty-six odor components were identified in the imported and domestic soybean samples.•When all 36 components were employed, the discrimination accuracy was mediocre, 83.0 %.•IAMB was employed to select the most descriptive components for discrimination.•When the three IAMB-selected components were used, the accuracy improved to 97.7 %. The identification of geographical origins of soybean pastes using headspace gas chromatography-mass spectrometry was attempted in this study. Since soybean paste was odor-rich, 36 components were identified in the imported and domestic soybean samples. t-Test, variable importance in projection (VIP), and Incremental Association Markov Blanket (IAMB) were employed to select proper components that could effectively discriminate the two sample groups. The discrimination accuracies were below 87.3 % when all 36 components were fed for either LDA, k-NN, or SVM. When the five t-test-selected components or six VIP score-selected components were employed, the accuracies improved to 95.2–96.2 %. The IAMB selected three different components were 3-methylbutanal, 4-methylnonane, and 2,3-pentanedione, and the correlations among their peak areas were not significant. This suggests that these three components were independently relevant for the discrimination. The accuracy obtained using these three components was superior, 97.7 %, as undescriptive and/or redundant components for the discrimination were excluded.