Hi, continuing on my analysis of RNAi screens, I have scaled and normalized my plates and am considering whether to perform a platewise variance adjustment (dividing by MAD(sample wells) on a given plate). When I do the calculation, I see that the standard deviation of the sample signals decreases when I perform the variance adjustment. However, the overall signal per plate (say as measured by the median) shifts dramatically. The result can be seen in a plate well series plot at http://rguha.net/varianceadjust.pdf where the upper plot is for the normalized data and the lower plot is the variance adjusted version of the normalized data. My question is: 1. While the variance across the plates has decreased by this adjustment, is there a good/reliable way to decide when to do it? 2. Why does the overall signal from each plate shift after the variance adjutsment? Does the result of the variance adjustment indicate that the overall signal on say plate 2 is lower than for plate 1? (Which is not apparent when just looking at the normalized data) Any pointers would be appreciated, -- Rajarshi Guha [[alternative HTML version deleted]]

Dear Rajarshi are you sure that you have the variance adjustment right? Simply dividing by the scale estimator (MAD) is not correct, you need to subtract the location estimator (e.g. median) first. Please check any statistics textbook on the topic of "z-scores" or "standardisation" for more on this. To decide whether the variance adjustment (or indeed, any other choice of preprocessing options) did well, please assess on the results from control wells, and from genes whose role in your assay of interest is known. Best wishes Wolfgang Rajarshi Guha wrote: > Hi, continuing on my analysis of RNAi screens, I have scaled and normalized > my plates and am considering whether to perform a platewise variance > adjustment (dividing by MAD(sample wells) on a given plate). > > When I do the calculation, I see that the standard deviation of the sample > signals decreases when I perform the variance adjustment. However, the > overall signal per plate (say as measured by the median) shifts > dramatically. The result can be seen in a plate well series plot at > http://rguha.net/varianceadjust.pdf where the upper plot is for the > normalized data and the lower plot is the variance adjusted version of the > normalized data. > > My question is: > > 1. While the variance across the plates has decreased by this adjustment, is > there a good/reliable way to decide when to do it? > 2. Why does the overall signal from each plate shift after the variance > adjutsment? Does the result of the variance adjustment indicate that the > overall signal on say plate 2 is lower than for plate 1? (Which is not > apparent when just looking at the normalized data) > > Any pointers would be appreciated, > -- Best wishes Wolfgang ------------------------------------------------------- Wolfgang Huber EMBL http://www.embl.de/research/units/genome_biology/huber