MARGE

MARGE models the relationship between H3K27ac ChIP-seq signals and gene expression to quantify regulatory potential and identify cis-regulatory elements in human and mouse genomes.

Key Features:

• MARGE-potential: Computes a regulatory potential (RP) per gene by summing H3K27ac ChIP-seq signals weighted by genomic distance from the transcription start site using a compendium of 365 human and 267 mouse datasets.
• MARGE-express: Uses logistic regression to select relevant H3K27ac profiles for modeling differentially expressed gene sets and retrieves signatures from 671 MSigDB gene sets to predict gene regulation, including BET-inhibitor repressed genes.
• MARGE-cistrome: Applies a semisupervised learning approach that integrates H3K27ac signal at DNase I hypersensitive sites from published human and mouse DNase-seq datasets to identify cis-regulatory elements and predict transcription factor binding without matched H3K27ac data.
• Comparison to superenhancers: Demonstrates improved prediction of BET-inhibitor repressed genes relative to superenhancer-based methods.

Scientific Applications:

• Transcriptional Regulation Studies: Integrates historical H3K27ac compendia with current experimental profiles to analyze mechanisms of transcriptional regulation.
• Cancer Research: Applied to RNA-seq and H3K27ac ChIP-seq from LNCaP-abl prostate cancer cells to study effects of silencing transcriptional and epigenetic regulators.
• Drug Response Prediction: Predicts BET-inhibitor repressed genes to inform drug-response studies and therapeutic research.

Methodology:

Computational methods explicitly include summing H3K27ac ChIP-seq signals weighted by distance to the transcription start site to compute regulatory potential; logistic regression to identify relevant H3K27ac profiles for differential gene sets using MSigDB gene sets; semisupervised learning to detect cis-regulatory elements; integration of H3K27ac signal at DNase I hypersensitive sites using published human and mouse DNase-seq datasets; and prediction of transcription factor binding in the absence of matched H3K27ac data.