Hi, I am working on the phenotypic screening of antiviral molecules. As a validation test for hits, it was suggested that I use the ‘Strictly Standardised Mean Difference’ (SSMD) analysis method, which does indeed appear to be more robust than a t-test or a Z-score. I perform the calculations using 96-well plates containing uninfected negative controls and infected controls. The first SSMD I calculate is the one that looks at the difference between the molecules and the infected positive controls. I consider a hit to be a molecule with an SSMD < -2. I then calculate another SSMD, this time looking at the difference between the molecules and the negative controls based on cell count. A good hit must have an SSMD greater than -1; otherwise, this indicates that it has cytotoxic properties. I then plot these two SSMD values on two axes and, using the thresholds given above, define a region where the hits are located for an experiment carried out in a 96-well plate. My question: I would like to group the results of several experiments onto a single graph. Is it necessary to include the negative and positive controls on the graph, given that the SSMD values obtained for the molecules are already normalised relative to these same controls? This question arises because I am unable to find any recent articles on this method, which seems to be little used yet is nevertheless interesting. Could you provide me with any recent references if you have any? Thanks François
1 answer
Yes, plot the controls, but use them as a per-plate QC axis rather than as decoration. A compound's SSMD is normalised against the controls on its own plate, so compound SSMDs from different plates are only on the same scale if each plate's control separation was comparable. Zhang's own QC framework does exactly this: compute the SSMD between positive and negative controls for each plate and treat that as the plate quality metric, much as you would use Z'. If one plate's control SSMD is noticeably weaker, its compound values aren't really poolable with the rest, and plotting them is how you see that.
On references, it's less that the method is little used and more that it lives in methods sections rather than standalone papers. The origin is Zhang 2007 in Genomics; the fullest treatment is his 2011 Cambridge book on optimal high-throughput screening design. For something free and actually maintained, the HTS assay validation chapters of the NIH Assay Guidance Manual on NCBI Bookshelf cover SSMD alongside Z'.
Log in to answer this question.