I am trying to manipulate my data set in cellHTS2 to apply a B-Score to my data and the scores that come out on my data report do not seem to match up with the values I would expect. I am working on compound screening and attempting to use the b-score to get rid of the edge effect on my data plates and show the best inhibitors from my screen and the data continues to come out without placing my strongest inhibitors as either the highest or lowest b-scores. The data comes out in a seemingly random order of b-scores, highly inhibitive compounds and those with almost no inhibition are appearing right next to each other in a random order when ranked highest to lowest based off of their b-scores. Has anyone else had similar problems or have any suggestions on how to fix this?
Thank you for the advice! I have just tried running the data set without normalizing or performing spacial corrections and it does not appear to fix my problem.
The data is visually represented in m-screen and shows rows near the edge of each plate where almost every compound falls into either the 20-40% or the 40-60% inhibition range and I am trying to come up with a way to see if any of those compounds are true positives or if they are all false positives as a result of the edge effect on the plate. The literature that I have read indicates that the b-score is a proven method to do this but I am up for other suggestions should you have any.
Also, I do not know what the locfit correction method is. Can you please send some more information on that.
Thank you again for all of you help!
For information about the locfit correction method please see the help documentation about normalization methods by typing ?normalizePlates in R. A lot of this comes down to the layout of compounds on the plate. If your assay is prone to spatial bias on the plates, it might be that you should not place key compounds or controls on the outermost edges and rows. Do you have strong inhibitors or positive controls on the interior of your plates? Do these behave as expected? If the spatial bias is extremely bad it will be a judgment call on your part as to whether or not to include the border wells at all in the analysis. It's good that you are thinking critically about the output of Bscore (and indeed any normalization method you try). This will help you to make a good decision.
I do not have any controls found on the inside of the plate, all of the controls on my 384 well plates are located in the two flanking columns on each side of the plate. I am not getting the highly inhibitive compounds to show up in a ranking order in the b-score analysis no matter where they are on the plate, along with not being able to parse the true from false positives on the edge of the plate. I am looking to find a successful control independent method that will allow me to pull out the true positives and rank them.
Are the controls in the border columns all the same, or do you expect to see a particular ranking due to e.g. a concentration gradient? At this point you're looking for any evidence that the experiment is indeed working as expected.
If your design is such that you require a high signal within the control wells to validate the experiment, but this is confounded with an increase of signal due to a spatial effect, then, if none of the spatial correction methods are helping, I would strongly recommend you repeat the experiment with a different layout for the controls.
If on the other hand such validation is not strictly necessary you could proceed just with the interior wells and ask which compounds are giving the strongest signal within each plate. After removing the affected border wells, using the "median" normalization method with varianceAdjust="byPlate" might still give you sensible results.