This repository contains the code for the manuscript 'A statistical workflow for the robust detection of interaction effects between chromatin modifications'.
We assume to be in a setting with an (binary) experimental design matrix L with n rows (observations) and q features. Under this design a corresponding multiple readout P with n rows and p columns is available. In our case L contains q different chromatin modifications and P contains the binding strength of p Proteins to the n nucleosomes. We are interested in the combinatorial (bi-order interaction) effect between those modifications on the binding behaviour of the proteins.
Our workflow includes the following main steps:
-
a Lasso for hierarchical interactions with a strong hierarchy (Bien et al., 2013) to reduce number of spurious interaction effects.
-
(Complementary pairs) stability selection for robust model selection in the data-scarce regime.
-
Replicate consistency check to remove noisy data that might cause spurious interaction effects.
Moreover, we provide a script (script_analysis_workflow.md) that shows how downstream analyses and visalization of the results from our workflow can be performed. In this script all figures in the manuscript can be reproduced.