Creating linked, interactive views of multivariate data

Visualizing data is an important component of exploratory data analysis. Ideally, one would be able to create views of the data, interact with those views to find interesting data points, and then perform statistical modeling of the data all within the same system. In the first paper, discussed in Chapter 1, two new R packages, MVCClass and iSPlot, are presented that allow users to create linked, interactive views of multivariate data. To create interactive views, two other R packages, RGtk and gtkDevice are used. Most importantly, the software was designed to be extensible using the model-view-controller paradigm to create linked views. This extensibility, which can be performed solely in the R language, lets developers create interactive views from the statically available plots in R and new responses to user interactions, such as changing the response to a left button click from coloring a point to hiding a point. In addition, since these packages are available in R, the user can perform statistical modeling in the same system as the exploratory data analysis.; In the second paper, discussed in Chapter 2, a generalization of the model-viewcontroller (MVC) paradigm is considered to allow for more flexible handling of submodels. Instead of using a single MVC a tree structure of related MVCs is developed, where each MVC has a single parent, but a parent can spawn many different child MVCs. Communication between parent and child MVCs allows for very flexible linking of the MVCs. An implementation of the work is provided in the three R packages, MVCClass, BIOMVCClass, and iSNetwork, available at http://www.bioconductor.org.; In the third paper, discussed in Chapter 3, the capabilities of the iSNetwork package are demonstrated by exploring Chronic Lymphocytic Leukemia (CLL) microarray data and its relationships with meta-data, such as the Gene Ontology (GO), the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, and the location of the genes on the chromosomes. Here, the data sets are large and complex, and simple pairwise scatter plots are not sufficient to explore the data. Using the iSNetwork package, more complex plots, such as heatmaps and graphs can be created by users. Being able to create views that are linked allows one to quickly explore multiple data sets jointly and simultaneously. However linking complex plots can be difficult and careful thought must go into how elements in one view are related to elements in another view, particularly when the elements are connected by a many-to-many relationship. The R package, iSNetwork, which was implemented using the design from the second paper, allows users to create linked, interactive views of linked data sets and thus, fulfills the goals of querying multiple data sets through linked, interactive views.