What is the standard practice to address plate layout artifacts in an HT drug screening: median polish won’t be work for a titration series or nonrandom plating. Also, if someone knows of an existing CellProfiler pipeline to address this, that will be a big help.
We typically do NOT address this in CellProfiler- features are measured on the images, and then normalized and corrected afterwards. In general, we do use median polishing, but a couple of other methods are discussed in this review (I’ve quoted the relevant section).
We recommend using a two-way median polish to correct for positional effects. This procedure involves iterative median smoothing of rows and columns to remove positional effects, then dividing each well value by the plate median absolute deviation to generate a B score60. However, this procedure cannot be used on nonrandom plate layouts such as compound titration series or controls placed along an entire row or column54. Other approaches include 2D polynomial regression and running averages, both of which correct spatial biases by using local smoothing61. Notably, image-based profiling is often sufficiently sensitive to distinguish among different well positions containing the same sample. Thus, to mitigate these positional effects, samples should be placed in random locations with respect to the plate layout. However, because such scrambling of positions is rarely practical, researchers must take special care to interpret results carefully and to consider the effects that plate-layout effects might have on the biological conclusions.
The absolute best thing to do though is to avoid this issue entirely; doing a “pseudo-random” plate design (where the order of series striping and/or placement changes- sometimes across a row, sometimes across a column, sometimes snaking and doubling back) can often help make it so that your samples are not always in the identical position but that you can plate the design without errors.
Hi Beth, how many such pseudo-random replicates do you suggest? Is there a recommended placement for them, since even this pseudo-random design may be biased?
For the primary screen, I can not do several replicates and thus need some statistical correction (unfortunately median polish did not work well for me). I will look into local smoothing.
For profiling experiments, we typically recommend a minimum of 4 replicates per conditions, 5 for safety.