Nonnegative matrix factorization in ASCA modeling of graphene oxide toxicometabolomics 1H-NMR data

Non-negative matrix factorization (NMF) has been shown to be a suitable decomposition method for multivariate data, to have a meaningful non-negativity physical interpretation [1]. The study is the employment of NMF to the calculation partitions variation and enables interpretation of these partitions in the ASCA model of metabolomics data set. ANOVA-Simultaneous Component Analysis (ASCA) is one of the most prominent methods to include such information in the quantitative analysis of multivariate data, especially when the number of variables is large [2]. The experimental dataset included 1HNMR spectra of mice serum samples exposed to three different doses (high, low and control) and at different time intervals such as 24 and 48 hours, 7 and 21days post-injection of GO nano-sheets [3]. In this data, to analyze the effects of the specific factors like time, dose and their interaction, an ASCA model based on NMF was applied. P-values with confidence interval 90% for factors low dose with control, low dose with high dose, control with high dose, time and interaction between time and dose were obtained 0.049, 0.214, 0.365, 0.001, and 0.997 respectively. As it was clear from the results, low with control group and time factor were effective factors in creating variance in the data. For the other two cases, the factor dose did not play an effective role in the process of data variances. Loading corresponding to the factor time and low dose with control showed the chemical shifts and thus the compounds that are corresponding to the behavior observed in the scores, and can be used for biological interpretation. Scores corresponding to the factor interaction no trend related to dose was visible and none of the four measurement time-points showed an increasing or decreasing score value for differing graphene oxide doses. Also the scores of the second component do not show such a quantitative trend. Considering the complexity of the metabolic data, the use of NMF instead of PCA in ASCA is an important and more realistic strategy to obtain meaningful physical profiles for each effective factor.