PARE: PARtial Embeddings for deconfounded dimension reduction

Maintainer: Andrew Chen, [email protected]

License: Artistic License 2.0

Partial embeddings (PAREs) provide a framework for removal of confounding effects from any distance-based dimension reduction method including $t$-SNE, UMAP, Laplacian Eigenmaps, diffusion map embeddings, and others. For Euclidean distances, PAREs are equivalent to applying dimension reduction to the principal coordinates after regressing out nuisance covariates.

1. Installation

The R package can be installed via devtools by running the following code

# install.packages("devtools")
devtools::install_github("andy1764/PARE", build_vignettes = FALSE)

Then, you can load this package via

library(PARE)

2. Usage

PARE leverages the existing R implementations for dimension reduction methods. Below is an example call for partial UMAP in the iris dataset, which removes the species effects:

library(PARE)
library(umap)

pe <- pare(iris[,-5], ~ Species, data = iris, umap)

A vignette is provided for the pare function, which also covers other dimension reduction methods. Once the suggested dependencies are installed, please run:

devtools::install_github("andy1764/PARE", build_vignettes = TRUE, force = TRUE)
vignette("pare")

3. Citations

If using the PARE methodology, please cite the following preprint:

Chen, A. A., Clark, K., Dewey, B., DuVal, A., Pellegrini, N., Nair, G., Jalkh, Y., Khalil, S., Zurawski, J., Calabresi, P., Reich, D., Bakshi, R., Shou, H., Shinohara, R. T., Initiative, the A. D. N., & Cooperative, the N. A. I. in M. S. (2023). Deconfounded Dimension Reduction via Partial Embeddings (p. 2023.01.10.523448). bioRxiv. https://doi.org/10.1101/2023.01.10.523448