Direct Classification of GC × GC-Analyzed Complex Mixtures Using Non-Negative Matrix Factorization-Based Feature Extraction

Complex chemical mixtures need to be evaluated to, for example, aid in medical diagnoses, assess product quality, and assess environmental conditions. Two-dimensional gas chromatography (GC × GC), which is a comprehensive analytical technique, combined with data classification techniques, has attracted great interest for assessing mixtures. In this study, a nontarget cross-sample analysis-based unsupervised direct classification method using non-negative matrix factorization was developed for assessing mixtures analyzed by GC × GC. The method was developed using GC × GC data for more than 30 river water samples as image data. The retention time shift correction data processing step was important to the classification accuracy because the compound signals were found at slightly different times for different samples. The maximum likelihood estimates of the matching ratios for the 30 samples, with retention time shift correction, were 86.8% and 77.0% using two and three ranks, respectively. The method is easy to perform and intuitive, requiring no specific knowledge or labeled data. This direct classification method will therefore be particularly useful for performing initial screens of large numbers of samples and for identifying major differences between samples.