HiPiler: Visual Exploration of Large Genome Interaction Matrices with Interactive Small Multiples

Fritz Lekschas, Benjamin Bach, Peter Kerpedjiev, Nils Gehlenborg, Hanspeter Pfister

Abstract

This paper presents an interactive visualization interface—HiPiler—for the exploration and visualization of regions-of-interest in large genome interaction matrices. Genome interaction matrices approximate the physical distance of pairs of regions on the genome to each other and can contain up to 3 million rows and columns with many sparse regions. Regions of interest (ROIs) can be defined, e.g., by sets of adjacent rows and columns, or by specific visual patterns in the matrix. However, traditional matrix aggregation or pan-and-zoom interfaces fail in supporting search, inspection, and comparison of ROIs in such large matrices. In HiPiler, ROIs are first-class objects, represented as thumbnail-like “snippets”. Snippets can be interactively explored and grouped or laid out automatically in scatterplots, or through dimension reduction methods. Snippets are linked to the entire navigable genome interaction matrix through brushing and linking. The design of HiPiler is based on a series of semi-structured interviews with 10 domain experts involved in the analysis and interpretation of genome interaction matrices. We describe six exploration tasks that are crucial for analysis of interaction matrices and demonstrate how HiPiler supports these tasks. We report on a user study with a series of data exploration sessions with domain experts to assess the usability of HiPiler as well as to demonstrate respective findings in the data.

Introduction

The human genome is a complex three-dimensional puzzle. While DNA is often represented as a linear sequence, it folds into intricate shapes within the cell nucleus, bringing distant genes into close physical contact. Biologists utilize Hi-C sequencing to generate "interaction matrices"—heatmaps that represent the probability of any two genomic locations interacting.

The primary challenge in analyzing this data is scale: these matrices can reach dimensions of 3 million by 3 million cells, totaling 9 trillion data points. Biologically significant patterns, such as "loops" or "domains," are essentially tiny needles in a massive haystack. Traditional visualization tools rely on a "pan-and-zoom" paradigm, which makes it nearly impossible to compare distant features or efficiently filter out the thousands of false positives generated by automated detection algorithms.

Methodology: The Snippet-Based Approach

HiPiler addresses these challenges by moving away from continuous image exploration toward a discrete, object-oriented workflow. The system decomposes the massive matrix into manageable "snippets."

The three-stage HiPiler workflow: extracting snippets from the matrix, organizing them, and interactive exploration.

Snippets as First-Class Objects

Instead of navigating the entire matrix, HiPiler extracts small square sub-matrices (ROIs) and displays them as interactive thumbnails. These snippets can be manipulated using several key features:

  • Attribute-Driven Layouts: Users can arrange snippets based on genomic metadata. They can be ordered linearly by metrics like "noisiness," plotted in 2D scatterplots (e.g., comparing distance from the diagonal against pattern sharpness), or clustered using t-SNE for multi-dimensional similarity analysis.
  • The "Piling" Metaphor: To manage hundreds of snippets, HiPiler allows users to "pile" them together. A pile displays a "cover" matrix representing the average or variance of all contained snippets, allowing researchers to identify consensus patterns while hiding outliers.
  • Brushing and Linking: The tool maintains a dual-view system. Selecting a snippet or pile in the gallery view instantly highlights its physical location in the global navigable matrix, providing essential spatial context.
The HiPiler interface featuring the global matrix view (left) and the interactive snippet gallery (right) for exploring regions of interest.

Results and Expert Evaluation

The authors evaluated HiPiler using high-resolution human genome interaction data (Rao et al., 2014) and ROI collections identified by algorithms like Delly and Meerkat. A qualitative study with genomic experts revealed several key insights:

  • Efficiency and Discovery: Experts noted that HiPiler enabled workflows that were previously too labor-intensive. Researchers were able to identify "non-centric" loops and correlate them with other genomic features far faster than with standard tools like HiGlass.
  • Noise Curation: Participants successfully used scatterplots and piling to rapidly filter out hundreds of low-quality snippets incorrectly flagged by automated algorithms. By looking at the variance cover of a pile, researchers could instantly spot which detected loops were consistent across the data.
  • Usability: Despite the data's complexity, experts found the interface intuitive, requiring only approximately 10 minutes of training to operate independently.

Conclusion and Future Work

HiPiler demonstrates that the "small multiples" approach is superior to single-image navigation for tasks involving categorization and outlier detection in genomic data. However, limitations remain. The current web-based implementation is optimized for roughly 2,000 snippets; beyond this, performance and visual clutter become significant hurdles. Future work will focus on supporting non-square (rectangular) ROIs, integrating real-time pattern detection directly into the interface, and refining biological normalization methods for aggregated piles.

Summary updated

Abstract

Abstract This paper presents an interactive visualization interface—HiPiler—for the exploration and visualization of regions-of-interest in large genome interaction matrices. Genome interaction matrices approximate the physical distance of pairs of regions on the genome to each other and can contain up to 3 million rows and columns with many sparse regions. Regions of interest (ROIs) can be defined, e.g., by sets of adjacent rows and columns, or by specific visual patterns in the matrix. However, traditional matrix aggregation or pan-and-zoom interfaces fail in supporting search, inspection, and comparison of ROIs in such large matrices. In HiPiler, ROIs are first-class objects, represented as thumbnail-like “snippets”. Snippets can be interactively explored and grouped or laid out automatically in scatterplots, or through dimension reduction methods. Snippets are linked to the entire navigable genome interaction matrix through brushing and linking. The design of HiPiler is based on a series of semi-structured interviews with 10 domain experts involved in the analysis and interpretation of genome interaction matrices. We describe six exploration tasks that are crucial for analysis of interaction matrices and demonstrate how HiPiler supports these tasks. We report on a user study with a series of data exploration sessions with domain experts to assess the usability of HiPiler as well as to demonstrate respective findings in the data.
BibTeX
@article{lekschas2018hipiler,
  author = {Fritz Lekschas and Benjamin Bach and Peter Kerpedjiev and Nils Gehlenborg and Hanspeter Pfister},
  title = {{HiPiler: Visual Exploration of Large Genome Interaction Matrices with Interactive Small Multiples}},
  journal = {IEEE Trans. Vis. Comput. Graph.},
  year = {2018},
  doi = {10.1109/tvcg.2017.2745978},
}
Figure