Multivariate Curve Resolution as a Pretreatment Step in Multivariate Analysis of Variance

In analytical chemistry, it is common to study the effect of changes in one or more factors (like temperature or type of catalyst) on a measured response for several possible purposes (like optimizing the yield of a reaction). Usually, in such studies, data is collected on the basis of an experimental design to guarantee that it will contain the information targeted by the study. Analysis of variance (ANOVA) is a statistical model that aims to decompose the total response variation to test the significance of each factor effect. ANOVA is utilized to determine the effect of the studied factors, as well as, to quantify the effect of different levels of each factor [1]. With the advancement of technology, it is possible to study the effect of the desired factors on multiple responses. As a result, ANOVA, due to its univariate nature, is not applicable to these datasets. ANOVA simultaneous component analysis (ASCA), which is an extension of ANOVA, is applicable in such cases [2]. In a dataset coming from an experimentally designed study, various chemical components can be influenced by the factors under investigation. ANOVA based methods decompose the dataset into matrices that represent the effects of each factor and their potential interactions, allowing for the examination of the effect of factors. However, in this decomposition, the effects of factors on the chemical components cannot be examined separately. Since the studied factors may have different effects on chemical components, investigating the effects of each factor on the chemical components can be crucial in various studies like metabolomics or food industry. MCR-ASCA approach has been proposed in this work to achieve the above-mentioned goal. In this approach, data is decomposed into the contributions of chemical components, and then the contribution of each component is decomposed with ASCA model for obtaining further information about the desired components. Several simulated and experimental data sets are applied for illustrating the advantages of applying this approach.