$30 off During Our Annual Pro Sale. View Details »

HiPiler: Visual Exploration Of Large Genome Interaction Matrices With Interactive Small Multiples

HiPiler: Visual Exploration Of Large Genome Interaction Matrices With Interactive Small Multiples

From my talk at InfoVis at IEEE VIS 2017 in Phoenix.

Fritz Lekschas

October 05, 2017
Tweet

More Decks by Fritz Lekschas

Other Decks in Science

Transcript

  1. HiPiler
    Visual Exploration of Large
    Genome Interaction Matrices with
    Interactive Small Multiples
    Fritz Lekschas, Benjamin Bach, Peter Kerpedjiev, Nils
    Gehlenborg, and Hanspeter Pfister

    View Slide

  2. 3 million × 3 million

    View Slide

  3. > 10.000 pattern instances but small total size

    View Slide

  4. > 10.000 pattern instances but small total size
    How can we
    explore and compare
    many local patterns in
    this very large matrix?

    View Slide

  5. HiPiler

    View Slide

  6. Social Networks
    Fans, connectors, and cliques
    Computer Networks
    Bottlenecks and hubs
    Gene Networks
    Feed-forward loops
    Giga-pixel Images
    Pattern recognition

    View Slide

  7. Structure of the Genome
    Acknowledgements: N. Abdennur, B. Alver, H.
    Belaghzal, A. van den Berg, J. Dekker, G. Fudenberg, J.
    Gibcus, A. Goloborodko, D. Gorkin, M. Imakaev, Y. Liu,
    L. Mirny, J. Nübler, P. Park, H. Strobelt, and S. Wang.

    View Slide

  8. DNA Cell Nucleus Contact Sequencing

    View Slide

  9. Cell Nucleus Contact Sequencing Matrix
    Fixed Ordering
    Altered DNA ordering is associated with severe diseases!

    View Slide

  10. Challenges
    • Detected by algorithms
    • Occur frequently
    • "Noisy" results
    Goals
    • Quality assessment
    • Pattern stratification
    • Pattern correlation
    Points Blocks

    View Slide

  11. • How do specific pattern or
    average pattern look?
    • How variant and noisy are
    detected pattern?
    • Are there subgroups among
    the pattern?
    • How are patterns related to
    other data attributes?
    • What does the patterns
    neighborhood look like?

    View Slide

  12. TECHNIQUES?
    • Pan & Zoom

    Kerpedjiev et al.: HiGlass
    • Lenses / Multifocus

    Rao and Card: Table Lense

    Elmquist et al.: Melange
    • Abstraction / Aggregation

    Dunne et al.: Motif Simplification

    Elmquist et al.: ZAME
    • Small Multiples

    Bach et al.: Multipiles

    View Slide

  13. Cut the Matrix into Pieces!

    View Slide

  14. Cut the Matrix into Pieces!

    View Slide

  15. Cut the Matrix into Pieces!

    View Slide

  16. Cut the Matrix into Pieces!

    View Slide

  17. Cut the Matrix into Pieces!

    View Slide

  18. Cut the Matrix into Pieces!

    View Slide

  19. HiPiler

    View Slide

  20. HiPiler

    View Slide

  21. 1. FILTERING
    Assess quality & separate signal from noise

    View Slide

  22. 1. FILTERING

    View Slide

  23. 1. FILTERING

    View Slide

  24. 1. FILTERING

    View Slide

  25. 1. FILTERING

    View Slide

  26. 1. FILTERING

    View Slide

  27. 2. AGGREGATE
    Stratify patterns and assess pattern variability

    View Slide

  28. 2. AGGREGATE

    View Slide

  29. 2. AGGREGATE

    View Slide

  30. 3. CONTEXT
    Correlate patterns with each another
    & other pattern types

    View Slide

  31. 3. CONTEXT

    View Slide

  32. Pile Inspection Attribute correlations
    Multidimensional Clustering Dataset Comparison
    More at http:/
    /vcg.seas.harvard.edu/pubs/hipiler

    View Slide

  33. User study with 5 domain experts: Evaluating usability and
    usefulness
    Snippet approach is useful: Average / variance assessment
    and parameter estimation
    Context matters: Coordination between the snippets and
    matrix is highly appreciated
    HiPiler is easy-to-use and useful: Domain experts ask for
    local installations
    Limitations: Fixed matrix ordering and fixed aspect ratio of
    snippets
    EVALUATION

    View Slide

  34. CONCLUSION
    Coordinate
    Aggregate
    Arrange &
    Filter
    Separate
    Explore

    View Slide

  35. NEXT?

    View Slide

  36. HiPiler
    PAPER

    vcg.seas.harvard.edu/pubs/hipiler
    LIVE

    hipiler.higlass.io
    CODE

    github.com/flekschas/hipiler

    View Slide