Detecting the presence of potential interactions between experimental factors are generally not straightforward in omics data analysis, especially for higher-order interactions (i.e. more than two factors). Several practical methods are available in MetaboAnalyst:
For individual variable, the interaction can be assessed by univariate two-way ANOVA or linear modeling (limma). Users can view a two-way box plot summary of each variable by clicking the corresponding name in the detailed table view of the two-way ANOVA results.
The overall interaction effect can be assessed by ASCA permutation tests (under the “Model Validation” tab on the ASCA page). It performs unrestricted permutation of observations (Manly’s approach) and recalculates the total sum of squares (TSS) for each experimental factor and their interactions. If the TSS calculated based on the original data is significantly different from those calculated from the permuted data, then the effect is significant. For more detailed discussions on this subject, please refer to the paper by Vis D. J. at. al
Through visualization approaches such as PCA, heatmap, etc to examine patterns of variations across different experimental factors