margins for points that have no matching value on the opposing axis
(Figure 4b). These gutter plots, like the columns in Figure 2, jitter
points in the unused direction to reveal density. Tooltips explain the
reason and the proportion of data in each area. Again, without a place
to show these data we would have to either filter them out or hide them
from view, which could prevent deeper understanding of the possible
lines of investigation. Like all parts of the plot, the gutters still
enable brushing and range selection that reveals other points from the
same subject anywhere else on the Plot.
Time axis
Comparing performance of an assay across multiple studies is a key
scenario. Time series views can help discover temporal patterns in
immune response or filter to data of interest, but vaccine studies
present special challenges. Every study has a different schedule. There
may be multiple vaccinations at different times with follow-up visits in
between and after. In addition, each study has multiple treatment groups
whose number and timing of events varies. Each group may also get
different vaccine products or doses. Finally, each assay may be run on
samples from different times than each other assay. Investigators must
know the key time points of a study in order to make meaningful
interpretations, but in the DataSpace they can see data from a dozen
studies at once. To address these challenges we created a custom time
axis in which every filtered study has a row of icons indicating when
any of its treatment groups had an event (Figure 5). There is a
different icon for vaccinations, follow-ups, and ”challenges,” when
animals are exposed to a virus directly to determine vaccine
effectiveness. Every icon can be brushed to highlight the associated
data in context and to show a tooltip explaining what products and doses
were given to each of the study’s groups with events at the time. Icons
without associated data for the current y-axis are still shown (but
desaturated) because their event timing can affect interpretation. For
example, investigators know to look 1-3 weeks from a vaccination to see
peak immune responses. The events can also be expanded to show a row of
icons for every treatment within the studies. Modifier keys can be added
to clicks to create filters like, ’day 357 for group one and day 252 for
group two,’ which might be comparable in meaning although separate in
time.
In analyses observed outside the DataSpace, we noted frequent
comparisons of groups or studies at the moment of peak immune response.
These moments may all be at different intervals from their