nd_scatter Interactive Biplot

Many systems biology studies derive datasets with hundreds or thousands of dimensions. For example micro-biome assays can produce relative abundance measurements for thousands of microorganisms and ATAC-seq (assay for accessible chromatin) or CHIP-seq (chromatin immuno precipitation) assays can produce gene accessibility and expression levels for hundreds of genes. This poster presents the nd_scatter widget -- an interactive visualization extending methods developed in GGobi [1] for exploring and presenting multidimensional data using three dimensional projections. The widget allows the user to select components, examine and adjust the projection vectors for the data, and to apply common projection method such as 3 dimensional principal components analysis or t-distributed Stochastic Neighbor projections. Mouse interactions rotate, pan, and zoom the 3 dimensional projection of the features on the 3 dimensional display. Data points of interest may also be identified and examined using a lasso tool. The widget provides an interactive and configurable implementation of a type of biplot [2]. The nd_scatter widget is a stand alone Javascript component built using HTML 5 canvas technology. It is based on the jp_doodle package ( https://github.com/AaronWatters/jp_doodle ) and it is designed to be easily embedded as an interactive Jupyter widget using jp_proxy_widgets ( https://github.com/AaronWatters/jp_proxy_widget ). The Jupyter notebook embedding is useful for including the widget visualization in scientific workflows or other computational narratives. The widget published as part of an open source project and is currently under active development. It is similar to the tensor flow embeddings tool [5] but can be used in any Javascript/HTML5 context.