| General Considerations |
 |
The methods used for analysis of high-throughput screening data
are as important as the screening protocols. There is no one correct
method, and different possibilities should be evaluated for individual
screens during the screen development process. Some general considerations
are discussed below.
Many assays involve a time-dependent readout and therefore have
background and intensity levels that vary over time and by plate.
Any screen with appreciable plate-to-plate changes in signal intensity
and background should first be scaled using fold induction by dividing
each observed well value by the plate median or the plate control-well
medians (depending on the experimental design). In general, using
the plate median is more reliable than the plate mean for re-scaling
or normalization as it is less affected by outlier values.
Screens without appreciable time- or plate-based signal intensity
variance should forego the fold-induction calculation and simply
be normalized on a plate-by-plate basis by calculating the z-score,
or number of standard deviations from the mean for each readout
value. These z-scores can be used to indicate the probability that
a screening positive is genuine and not due to background noise.
As mentioned earlier, duplicate data points are very important
for determining which compounds are genuine positives meriting
follow-up. As an example, consider one pair of duplicate data points
with z-scores of 0.5 and 2.0. The data point with a z-score of
0.5 represents an event that is about 61.7% probable to have occurred
randomly, whereas the data point with a z-score of 2.0 represents
an event with a 4.5% chance of having occurred randomly. Averaging
these probabilities gives a 33.1% chance of random occurrence.
In comparison, a data point with a z-score of 1.5 in both duplicates
has a 13.4% chance of random occurrence in both cases, making it
a better screening positive to follow up.
|