Run a Differential Expression - Sliding Window ANalysis (DEswan) as described by Lehallier et al. (https://www.biorxiv.org/content/10.1101/751115v1)

Installation

You can install DEswan from github with:

# install.packages("devtools")
devtools::install_github("lehallib/DEswan",build_vignettes = T)

DEswan approach

DEswan explore linear and non linear relationships between a quantitative trait (l) and one or more features.

Considering a vector l of k unique values, we iteratively use lk as the center of a window of size x and group samples in parcels below and above lk

i.e. [lk-x/2 ; lk[ and ]lk ; lk+x/2]

To test for differential expression, we use the following linear model:

Featurei ~ α+β1 lk(below/above)+ε

lk(below-above) being binarized according to the parcels around lk. Covariates can be included in the modeling as follows:

Featurei ~ α+β1 lk(Low/High)+β2 Covariatea+…+βx Covariatex

Type II sum of squares are calculated using the Anova function of the R car package.

When analyzing the links between l and more than one feature, we recommend to estimate q-values for each lk using Benjamini–Hochberg correction. To assess the robustness and relevance of DE-SWAN results, we recommend to test multiple parcel widths and different p/q-values thresholds.

DEswan example

See the vignette for examples