Explorative visual analysis of meta-’omics data requires multivariate visualization methods that accommodate high-dimensional hierarchical datasets while balancing aggregate patterns with sample-level detail. Current approaches typically emphasize either aggregate summaries or detailed views, but rarely integrate both perspectives effectively for ’first-look’ analysis without requiring consequential pre-exploration decisions about dimensionality reduction or hierarchical level selection.
This PhD project investigates multivariate visualization for high-dimensional hierarchical data through four contributions addressing effectiveness, scaling, and hierarchical integration. Comparative evaluation of the Scatter Plot Matrix (SPLOM) and Parallel Coordinates Plot Matrix (PCPM) across foundational tasks reveals no substantial performance differences for outlier detection, though strong user preference for the PCPM suggests value in providing method choice. Empirical analysis of Parallel Coordinates Plot scaling demonstrates that aspect ratio systematically affects task performance: ratios greater than 1 improve positive correlation perception but extreme ratios degrade value tracing accuracy. The novel Icicle Parallel Coordinates Plot (IPCP) exploits hierarchical structure to provide scalable overview+detail visualization by combining icicle plots for aggregate representation with parallel coordinates for rendering sample-level relationships across multiple levels simultaneously. Qualitative evaluation with domain experts demonstrates effective navigation, outlier identification, and balance between overview and detail, with potential for transferability to network traffic analysis and other domains.
Team
- Hugh Garner (PhD student), Newcastle University
- Dr Sara Johansson Fernstad (main supervisor), Newcastle University
- Dr Carmen Montero-Calasanz (co-supervisor), formerly Newcastle University
- Prof Nick Holliman (co-supervisor), formerly Newcastle University
- Prof Anil Wipat (co-supervisor), Newcastle University