Epilogos: information-theoretic navigation of multi-tissue functional genomic annotations

Jimmie E. Quon, Alex Reynolds, Nalu Tripician, Eric Rynes, Athanasios Teodosiadis, Manolis Kellis, Wouter MeulemanbioRxiv 2025 · DOI

TL;DR

This paper introduces Epilogos, a novel information-theoretic framework for navigating and interpreting complex multi-tissue functional genomic annotations, particularly chromatin state maps. By quantifying the "saliency" of genomic regions, Epilogos enables researchers to discover biologically relevant patterns, compare different cell types or conditions, and search for similar genomic regions across the entire genome, all through an intuitive web browser.

What Problem Does This Paper Address?

The human genome is incredibly complex, and functional genomics data, such as chromatin state maps, offer deep insights into biological processes. However, the sheer volume and dimensionality of this data (thousands of cell types, tissues, and experimental conditions) make it incredibly difficult to navigate, interpret, and find meaningful patterns. Existing tools often struggle to handle this scale, leaving a gap in our ability to discover commonalities, differences, and prioritize regions of interest across large datasets. This paper aims to bridge that gap by providing a scalable and intuitive framework for exploring this rich genomic landscape.

Methods

Epilogos leverages principles from information theory, drawing an analogy to sequence logos used for DNA or protein motifs. The core innovation is the concept of "saliency," which quantifies how informative or unexpected a particular chromatin state is at a given genomic position, relative to genome-wide expectations. They define three saliency metrics (S1, S2, S3) that progressively incorporate more complex background models, from simple label frequency to pairwise state co-occurrence and even biosample relationships.

Key methodological contributions include:

  • Saliency Scores: Quantifying the informativeness of genomic regions based on their chromatin state patterns across multiple biosamples.
  • Consensus Epigenome: Constructing a representative epigenome by selecting the most salient state at each position, which can highlight rare but important states missed by simple averaging.
  • Differential Analysis: Enabling direct pairwise comparisons between groups of biosamples (e.g., male vs. female, cancer vs. non-cancer) to identify significant epigenomic differences.
  • Similarity Search: A novel approach to find genomic regions with similar epigenomic patterns, even with slight variations in alignment or spacing, by using a scale-normalized block-based representation.
  • Web Browser and Open-Source Tools: A user-friendly web interface (http://epilogos.net) and accompanying open-source software for broad accessibility and adoption.

Datasets

The Epilogos framework was applied to several large-scale, publicly available datasets:

  • Roadmap Epigenomics Project: 127 human epigenomes with 15-, 18-, and 25-state chromatin annotations.
  • EpiMap project: 833 human epigenomes with 18-state annotations.
  • International Human Epigenome Consortium (IHEC): 1,698 human epigenomes with 18-state annotations.
  • Mouse epigenomes: 65 mouse epigenomes with 15-state annotations.

These datasets represent a significant portion of the available genome-wide chromatin state data, encompassing over 2,000 maps in total.

Key Insights and Ideas

  • Saliency as a Universal Metric: The concept of saliency provides a data-driven way to prioritize genomic regions, independent of pre-defined annotations, and has been shown to correlate with evolutionary conservation, trait-associated genetic variation, and regulatory potential.
  • Beyond Averaging: The consensus epigenome derived from maximal saliency states can reveal functionally important, rare chromatin states that might be obscured in simple average-based representations.
  • Unbiased Differential Analysis: Epilogos's approach to differential analysis is fully data-driven, not relying on pre-defined genomic loci or states, allowing for the discovery of novel regulatory regions.
  • Visual Intuition for Similarity: The similarity search leverages human visual intuition, recognizing that functional similarity in epigenomic patterns might exist even with minor structural variations, making it a powerful tool for exploration.
  • "Augmented Genomics": The paper frames Epilogos as part of an emerging paradigm of "augmented genomics," combining large-scale visualization with data-driven inference to help scientists navigate the ever-growing complexity of genomic data.

Key Results

  • Epilogos successfully highlights well-characterized and previously underappreciated genomic regions, including non-coding and potentially novel regulatory loci.
  • Saliency scores strongly correlate with evolutionary sequence conservation (Fig. 1e), enrichment of genetic variants with strong trait associations (Fig. 1f), and regulatory sequence content (Fig. 1g).
  • Pairwise comparisons revealed significant epigenomic differences, such as those related to X-inactivation between male and female donors, and identified key regulatory loci associated with cell cycle and development when comparing stem and non-stem cells.
  • The similarity search successfully identified regions with similar promoter architectures and enhancer clusters across the genome.
  • The Epilogos browser provides a fast, lightweight, and interactive platform for exploring over 2,000 chromatin state maps.

Limitations and Future Work

  • Computational Complexity: While Epilogos is designed for scalability, the S3 saliency metric, which incorporates biosample-specific relationships, incurs greater computational complexity.
  • Generalizability: While the framework is broadly applicable to other genome-wide annotations (e.g., gene annotations, evolutionary conservation), the current implementation and validation focus on chromatin state segmentations.
  • Future Work: The authors suggest that Epilogos can serve as a framework to integrate and refine existing or future methods for quantifying pairwise differences. The open-source nature of the project encourages community adoption and further development. The "augmented genomics" paradigm itself suggests a future direction for integrating diverse data types and analytical approaches.

Summary updated

Abstract

Abstract Functional genomics data, such as chromatin state maps, provide critical insights into biological processes, but are hard to navigate and interpret. We present Epilogos to address this challenge by offering a simple information-theoretic framework for large-scale visualization, navigation and interpretation of functional genomics annotations, and apply it to over 2,000 genome-wide chromatin state maps in human and mouse. We construct intuitive visualizations of multi-tissue chromatin state maps, prioritize salient genomic regions, identify group-wise differential regions, and enable rapid similarity search given a region of interest. To facilitate usability, we provide a purpose-built web-based browser interface ( http://epilogos.net ) alongside open-source software for community access and adoption.
BibTeX
@misc{quon2025epilogos,
  author = {Jimmie E. Quon and Alex Reynolds and Nalu Tripician and Eric Rynes and Athanasios Teodosiadis and Manolis Kellis and Wouter Meuleman},
  title = {{Epilogos: information-theoretic navigation of multi-tissue functional genomic annotations}},
  year = {2025},
  doi = {10.1101/2025.06.18.660301},
  howpublished = {bioRxiv},
}

Loading related content...

Figure