Sample Validity Platform - Details
Variables:
Spikes in the measurement - This value has the units of # of standard deviations away from the mean (listed as "X" in test #1 below).
Drifting measurements over time - measured in a fractional change (listed as "TT" in Tests #2 and 3c below) -- a value of 0.1 indicates that if the parameter values drift more than 10% over the collection time, it's invalid.
Uniform event acquisition - expressed as event rate over the average that any 2% of the events have to be in order to be invalid (listed as "Rt" in Test #3b below).
FlowJo (V8 Mac) does up to five different consistency checks on the data (the last three fall into the category of event rate checking). For most of the calculations, FJ divides the data files into "n" fractions, where n is a minimum of 7 and a maximum of 40; n scales roughly logarithmically with total number of events. It then computes, for every collected parameter, the median value for that parameter for each fraction of events, 1-n.
Each of the checks below is done for each collected parameter; described is the check done on a single parameter.
(1) Consistency of parameter values across the collection.
Invalid if: Abs(Mi - Mm) > X * SDm
Mi = the median value for fraction #i (out of n). Mm = mean of the "n" median values; SDm = the standard deviation of the "n" median values.
X = the user specified value for the kinetics threshold. Units are in # of standard deviations; i.e., it is like a T-score. A value of 4 triggers an invalidity when a single slice's median value is more than 4 SD away from the mean for all of the medians for that parameter.
Basically, this tests to make sure the median for a parameter for any given time slice is not much different than the medians for all other slices.
(2) Consistency of parameter values over time.
Invalid if: Sm x n / Mm > TT
Sm = the slope of the medians across the n fractions; n = # of fractions; Mm = mean of the medians. In other words, the value on the left side is a unitless slope that calculates to the same value irrespective of absolute parameter scaling or the number of fractions. A value of 1 for this ratio would mean that the median of the last fraction is twice as great as the median of the first fraction (i.e., the mean is halfway between and the range is equal to the mean).
TT = the user specified value for trend threshold.
(3) Uniform acquisition rate.
There are three tests for the uniform acqusition rate:
(3a) Invalid if there is a significant time block without events collected
Invalid if: dT for any consecutive events is > 50 x MdT
dT is the time between two events; MdT is the average time between two events for the entire collection (excluding "zero" time between events, i.e., the resolution on the time parameter is not sufficient to distinguish the time). There are no user-specified values for this test.
(3b) Invalid if there is a cluster#$&@ of events.
A cluster#$&@ of events is defined as when, at any point, the time required to collect any 2% of consecutive events in the file is either less than
MdT / Rt
OR if the time for 2% of the event is greater than
MdT * Rt
Where MdT is the average time between events, and Rt is the user-specified value for the rate threshold.
Basically, this makes sure there are no significant numbers of events collected either way too fast or way too slow.
(3c) Invalid if there is a significant trend in the rate of events collected
For each of the "n" fractions, the average rate of events for that slice is computed. The same test as #2 above is applied, i.e.,
Invalid if: Sr x n / Mr > TT
Sm = the slope of the rates across the n fractions; n = # of fractions; Mr = mean of the rates.
TT = the user specified value for trend threshold (same as #2 above).
