rr: (R)obust (R)e-scaling to help reveal latent structure.

The Robust re-scaling transformation (RR) is a transformation the help reveal latent structure in data. It uses three steps to transform the data:

1. Gaussianize the data,
2. z-score Transform the data,
3. remove extreme outliers from the data.

The sequence of these transformations helps focus analyses on consequential variance in the data rather than having it be focused on variation resulting from the feature’s measurement scale or outliers.

Software

• Our R package rrscale is available on CRAN: https://cran.r-project.org/package=rrscale on as part of this github repo.

Vignettes

• A simple introduction to rrscale is available here (html rmd)

Papers

1. “Automatic Transformation and Integration to Improve Visualization and Discovery of Latent Effects in Imaging Data” JCGS (2020)

• Our publication in JCGS is accessible here

• Our docker image for reproducing the analysis in paper can be found here.

• The scripts for the analysis in our paper can be found here

• The data for for paper can be found as part of our docker image, on github here, or at the MEP-LINCS synapse page here