Iterative Design and Evaluation of Regulatory Network Visualisation at Scale.

Over the last decade, the development of a range of Next Generation Sequencing (NGS) technologies has led to an enormous increase in the size of the data sets available in molecular biology. The scale of these data presents new challenges for researchers, and visualisation is widely regarded as an essential tool for exploration and detailed analysis of candidate relationships. Inevitably, there are cognitive and technical limits on the information which may usefully be displayed on a particular device, and there may be some tension between the analytical utility of a representation and its coverage of the relationships available within the data. Careful attention must be given to the overall design of the visualisation, and to the channels selected, and these tasks are further complicated if the intent is to support interactive exploration by a number of collocated researchers or inclusion within a collaborative workflow. This paper is concerned with the design of a visualisation for regulatory interactions in bacteria, the complex relationships that exist between a set of proteins and the much larger set of genes whose action they control. Modelling these interactions yields equally complex network diagrams, and even classical hairball representations when visualised. In this work we explore the iterative refinement of an alternative visualisation for data of this kind, moving away from the traditional hairball to a u0027fieldu0027 of smaller structures, the intent being to support effective comparison across many dozens of strains and species rather than the exhaustive documentation of a full set of interactions for the one organism. While the study did not directly compare insights obtained using TRNDiff with those obtained using other tools, formal evaluations have allowed us to settle on an effective set of representations and visual channels, and interactive features to support analysis. Our approach has produced a far more effective visualisation of these important data sets, and offers useful lessons for tool developers and insights into the utility of touch devices and larger displays for visual analytics and generation of insight at scale.