Simultaneous detection and estimation of trait associations with genomic phenotypes

TL;DR

This paper introduces JADE, a new method for comparing dense genomic phenotypes across groups by leveraging spatial structure to estimate profiles and detect differential regions, demonstrated through simulations and real data.

Contribution

The paper presents JADE, a novel approach that simultaneously estimates smooth genomic profiles and identifies differential regions, improving analysis of spatially structured phenotypes.

Findings

JADE effectively detects differential methylation regions.

Simulation results show high accuracy in profile estimation.

Application to muscle cell data reveals biologically relevant differences.

Abstract

Genomic phenotypes, such as DNA methylation and chromatin accessibility, can be used to characterize the transcriptional and regulatory activity of DNA within a cell. Recent technological advances have made it possible to measure such phenotypes very densely. This density often results in spatial structure, in the sense that measurements at nearby sites are very similar. In this paper, we consider the task of comparing genomic phenotypes across experimental conditions, cell types, or disease subgroups. We propose a new method, Joint Adaptive Differential Estimation (JADE), which leverages the spatial structure inherent to genomic phenotypes. JADE simultaneously estimates smooth underlying group average genomic phenotype profiles, and detects regions in which the average profile differs between groups. We evaluate JADE's performance in several biologically plausible simulation settings. We also consider an application to the detection of regions with differential methylation between mature skeletal muscle cells, myotubes and myoblasts.