What-if effects: A tale of entangled covariates.

Yu, Zhaowei; Zhang, Yong · Patterns (N Y) · 2026

Where this comes from

Abstract

Single-cell epigenomic studies promise to reveal how disease states, donor traits, environmental exposures, and technical factors shape cellular regulatory programs, yet these variables are often highly collinear and difficult to disentangle, particularly in human cohorts. The recent study by Møller and Madsen in <i>Patterns</i> introduces DeepDive, a probabilistic deep-learning framework that disentangles known covariate effects and residual variations in single-nucleus ATAC-seq data, enabling counterfactual "what-if" analyses of chromatin accessibility.