Multivariate Methods with Unbiased Variable Selection

Predictive multivariate modelling for metabolomics. Types: Classification and regression. Methods: Partial Least Squares, Random Forest ans Elastic Net Data structures: Paired and unpaired Validation: repeated double cross-validation (Westerhuis et al. (2008), Filzmoser et al. (2009)) Variable selection: Performed internally, through tuning in the inner cross-validation loop.

MUVR2

Multivariate methods with Unbiased Variable selection in R
PhD candidate Yingxiao Yan yingxiao@chalmers.se
Associate Professor Carl Brunius carl.brunius@chalmers.se
Department of Life Sciences, Chalmers University of Technology www.chalmers.se

General description

The MUVR package allows for predictive multivariate modelling with minimally biased variable selection incorporated into a repeated double cross-validation framework. The MUVR procedure simultaneously produces both minimal-optimal and all-relevant variable selections.

The MUVR2 package is developed with new functionalities based on the MUVR package.

An easy-to-follow tutorial on how to use the MUVR2 package can be found at this repository at inst/Tutorial/MUVR_Tutorial.docx

In brief, MUVR2 proved the following functionality:

• Types: classification, regression and multilevel.
• Model cores: PLS, Random Forest, Elastic Net.
• Validation: repeated double cross-validation (rdCV; Westerhuis et al. 2008, Filzmoser et al. 2009).
• Variable selection: recursive feature elimination embedded in the rdCV loop.
• Resampling tests and permutation tests: assessment of modelling fitnness and overfitting.

Installation

• You will need to have installed R (https://www.r-project.org/)
• Normally, you will want to work in RStudio (https://rstudio.com/) or some other IDE

You also need to have the remotes R package installed. Just run the following from an R script or type it directly at the R console (normally the lower left window in RStudio):

install.packages('remotes')

When remotes is installed, you can install the MUVR2 package by running:

library(remotes)
install_github('MetaboComp/MUVR2')

References

• Yan Y, Schillemans T, Skantze V, Brunius C. Adjusting for covariates and assessing modeling fitness in machine learning using MUVR2. Bioinformatics Advances. 2024, 4(1), vbae051.
• Shi L, Westerhuis JA, Rosén J, Landberg R, Brunius C. Variable selection and validation in multivariate modelling. Bioinformatics. 2019, 35(6), 972–80.
• Filzmoser P, Liebmann B, Varmuza K. Repeated double cross validation. Journal of Chemometrics. 2009, 23(4), 160-171.
• Westerhuis JA, Hoefsloot HCJ, Smit S, Vis DJ, Smilde AK, Velzen EJJ, Duijnhoven JPM, Dorsten FA. Assessment of PLSDA cross validation. Metabolomics. 2008, 4(1), 81-89.