lmdme: Linear Models on Designed Multivariate Experiments in R

Abstract

Thelmdmepackage decomposes analysis of variance (ANOVA) through linear mod-els on designed multivariate experiments, allowing ANOVA-principal component analysis(APCA) and ANOVA-simultaneous component analysis (ASCA) inR. It also extends bothmethods with the application of partial least squares (PLS) through the specification ofa desired output matrix. The package is freely available fromBioconductorand licensedunder the GNU General Public License.ANOVA decomposition methods for designed multivariate experiments are becomingpopular in “omics” experiments (transcriptomics, metabolomics, etc.), where measure-ments are performed according to a predefined experimental design, with several exper-imental factors or including subject-specific clinical covariates, such as those present incurrent clinical genomic studies. ANOVA-PCA and ASCA are well-suited methods forstudying interaction patterns on multidimensional datasets. However, currently anRimplementation of APCA is only available forSpectradata in theChemoSpecpackage,whereas ASCA is based on average calculations on the indices of up to three design ma-trices. Thus, no statistical inference on estimated effects is provided. Moreover, ASCA isnot available in anRpackage.Here, we present anRimplementation for ANOVA decomposition with PCA/PLSanalysis that allows the user to specify (through a flexibleformulainterface), almostany linear model with the associated inference on the estimated effects, as well as todisplay functions to explore results both of PCA and PLS. We describe the model, itsimplementation and two high-throughputmicroarrayexamples: one applied to interactionpattern analysis and the other to quality assessment.