Application
\label{application}
Currently, 17 studies with 570 humans and non-human primates have 84
dimensions to explore, and more studies and assays are in the pipeline.
Treatments include placebos, active controls, “challenges” to animals
in which they are exposed to an infectious HIV-like virus, and 35
vaccines and adjuvants on varied schedules of administration. Four
assays provide measures of immune system response to these treatments
based on testing subject samples against antigens: the binding antibody
multiplex assay (4 antibody isotypes and 19 antigens), ELISpot (IFNγ for
14 antigens), intracellular cytokine staining (2-cytokine response for 2
cell types against 13 antigens), and neutralizing antibody (3 target
cells and 89 antigens). In some cases, we combined results from multiple
labs running the same standardized assay. During prototyping, we
demonstrated scalability to hundreds of real studies with thousands of
subjects and we plan to expand to that scale again.
The DataSpace homepage has usage tips, news, and access to previously
saved groups and plots. The Learn section is a unique source of
extensive metadata for hundreds of studies, vaccine products, and
assays, hyperlinked by association. Find Subjects is for filtering
cohorts by categorical values, Plot supports up to 3 variables, and the
Data Grid shows a spreadsheet view for filter or export. A session
begins with all data available. Filters made in one place affect every
other data view. We describe select visual analytics capabilities in
more detail below.
Info Pane
\label{info-pane}
A persistent pane on the right lists active filters and lets
investigators track the key data attributes that the filters affect.
Without this list, it is easy to change a plot or filter without
realizing ancillary effects like removing studies or skewing the subject
population. At times, these counts either might be candidate
explanations of a visual pattern or confounds to another explanation.
Clicking a count shows the current filtered values and allows direct
change. We worked with investigators to prioritize the most important
counts, including subjects, species, studies, products, treatments, and
time points (Figure 2).
Find Subjects
\label{find-subjects}
This section has a bar chart counting the number of subjects in the
DataSpace with different attributes, such as their species or the
vaccines they received. Like the cohort selection from Bernard et al.,
brushing shows overlap in other values, enabling, for example, seeing
how many subjects were additionally given any other product (such as a
complementary vaccine or an adjuvant) (Figure 1). Filtering with a bar
value removes all the subjects without the attribute and keeps the
subjects that have overlapping values. Filters affect all other data
views in the application. Controls at the top for changing the attribute
and its hierarchical sorting allow discovery of attribute relationships
in available data and serial filtering such as, ’all macaques that
received any kind of DNA vaccine that were tested on the neutralizing
antibody assay.’ Unlike Bernard et al., we