Metabolomics Analysis of Metabolite Differences during the Storage Process of Laba Garlic Based on Machine Learning

Abstract

In this study, non-targeted metabolomic research was conducted using gas chromatography-mass spectrometry (GC-MS) on Laba garlic at different storage times to explore the signature metabolite changes during storage. Multivariate statistical analysis was used to screen out the significant metabolites of Laba garlic at different storage times. Further, the deep learning support vector machine recursive feature elimination (SVM-RFE) algorithm was used to determine its landmark metabolites, and enrichment analysis of the pathways of its landmark metabolites was carried out. A total of 57 different metabolites were screened out (VIP≥1, P<0.05, FDR<0.05) based on partial least squares-discriminant analysis (PLS-DA) as a criterion for screening, including acids (6), alcohols (12), amines (6), glycosides (1), esters (3), ethers (4), aromatics (13), olefins (2), and others (10). Further, 6 landmark metabolites were screened based on the SVM-RFE process, among which 5-hexyn-1-ol was the key factor to distinguish garlic and Laba garlic, while the organic active small molecule including inositol, 9-octadecenamide, stearic acid, palmitic acid and benzoic acid were the key factors to distinguish the 35th and 85th days of Laba garlic storage. KEGG enrichment analysis indicated that the more important metabolic pathways during storage were biosynthesis of unsaturated fatty acid, fatty acid biosynthesis, cutin, suberine and wax biosynthesis, ascorbate aldarate metabbolism, fatty acid elongation, phosphatidylinositol signaling system, galactose metabolism, inositol phosphate metabolism and fatty acid degradation. The six landmark metabolites in the storage process of Laba garlic analyzed in this study would provide a theoretical basis for the storage and evaluation of Laba garlic.

Keywords

• metabolomics
• laba garlic
• machine learning
• landmark metabolites
• support vector machine (svm)