Covariate-Sensitive Analysis of Cross-Sectional High-Dimensional Data

Using non-parametric tests, naive associations between omics features and metadata in cross-sectional data-sets are detected. In a second step, confounding effects between metadata associated to the same omics feature are detected and labeled using nested post-hoc model comparison tests, as first described in Forslund, Chakaroun, Zimmermann-Kogadeeva, et al. (2021) . The generated output can be graphically summarized using the built-in plotting function.

metadeconfoundR

CRAN/METACRAN License

metadeconfoundR was developed to perform a confounder-aware biomarker search of cross-sectional multi-omics medical datasets. It first detects significant associations between individual supplied features and available metadata, using simple nonparametric tests like mann whitney u test. In a second step, potential confounding effects between different metadata variables are detected, using nested linear model comparison post-hoc tests.

metadeconfoundR is also able to incorporate prior knowledge about confounding effects into this second analysis step. Drug association knowledge gained and reported from analyses of the MetaCardis cohort (Forslund et al., 2021) could, for example, now be used as additional input for future studies encompassing the same omics modalities and available metadata. Now, known confounders will be treated as such even if statistical power in the new dataset is not sufficient to detect them, thereby reducing the risk of drawing wrong conclusions based on undetected confounders. Details about this can be found in the latest release notes of metadeconfoundR.

Instalation

# in R
# from CRAN:
install.packages("metadeconfoundR")
# developmental version from GitHub:
library(devtools)
install_github("TillBirkner/metadeconfoundR@develop")