Related work
\label{related-work}
The Atlas repository for HIV vaccine studies supports plotting at the
level of a dataset, which is typically a single assay from a single
study [4]. The Neutralizing Antibody tool in LabKey Server supports
more detailed data management and analysis than the DataSpace, but for a
single assay [5]. Al-Hajj et al. and Shih et al. used Tableau for
analysis of similar immune system data and found evidence of the value
of visual analytics [6][7].
Physicians studying medical data for insights about patients have
similar VA tasks. LifeLines 2 and the work of Bernard et al. are salient
examples that share design patterns with the DataSpace for cohort
selection and time comparison [8][9]. This area of research is
based on electronic health record data where individual cases vary
widely in categorical attributes and event data, whereas subjects in
vaccine studies adhere to study selection criteria and follow
randomized, controlled protocols designed to enable planned analysis of
multiple quantitative metrics.
Al-Hajj, Shih, Bernard, and LifeLines 2 used paired analytics, requiring
a tool expert to work with a domain “user.” We aim for self-service
and scale, which has led us to do more data integration in advance,
provide more annotation, and focus on usability for the most common
tasks.
TrialShare, ImmPort, and ImmuneSpace are most similar to the DataSpace
in their goals to make studies broadly available with tools for analysis
[10][11][8][12]. In some ways, they are more ambitious
than the DataSpace: they support user scripts for reproducible analysis
and include datasets from over a dozen immune conditions, other study
types beyond vaccination, and more diverse assays. However, their
built-in visualizations are mostly static, and currently work best with
single studies or single assays across studies. In comparison, the
DataSpace offers deeper interaction for a narrower set of tasks. While
it is less flexible than code scripts, our graphical approach is meant
to lower the barriers to quick insights. We believe these approaches are
complementary and may benefit from each other in the future.
Many investigators in our work use a variety of general-purpose (R, JMP,
Prism, Excel) and assay-specific (FlowJo) visual tools for their own
data. Once an investigator identifies a cohort or idea in the DataSpace,
these external tools will remain critical for deeper exploration or
statistical validation.