Resolution:
## Figure 10.
Graphical quality control limits. The full population of triplicate sets shown in Figure 9 is here represented as
a histogram of the individual standard deviations for velocities from triplicate wells
of KM101 cells (black bars) and MG-63 cells (white bars). Presented in this fashion,
the data are useful for graphically estimating quality control limits and identifying
outliers. Data from images of MG-63 cells are shown to be "well behaved" in comparison
to KM101 cells, which exhibit a number of outliers to the right of the "normal" limit
expected for the tail. Arrows indicate positions of "3 sigma" limits that are based
on the observed percentage (99.7%) of observations (rather than on a more complex
transformation of the data according to a mathematical model). In the case of MG-63
cells, the limit (block arrow) excludes several adjacent triplicate sets that are
properly part of the distribution. Two certified MG-63 outliers are not readily visible
on this chart at 0.40–0.41 um/min. In the case of KM101 cells, the "3-sigma" limit
(solid arrow) more haphazardly cuts off a chain of outliers. As described in the text,
one source of KM101 outliers was identified by manual viewing of tracked objects in
the image sequences, and may lead to a rational filtering method that would improve
the data on both sides of the control limit for KM101 cells. Since the average variance
for triplicate velocity measurements correlated to some extent with average velocity
(see Figure 9), a more refined approach to quality control would involve first ranking
the data according to average velocity. Presumably the next step would be to determine
whether a single member of the outlying triplicate sets should be discarded in favour
of preserving the remaining two measurements in each case.
Bahnson |