In my smallest couple of standards on the standard curve for RT-qPCR, there are a couple of wells that don't quite fit the standard curve very well--they look like outliers, but how do you determine whether a standard well is an outlier or not? Is there a protocol for this? I always delete the standard wells that never met Ct and the standard wells that look far off from the curve, but I want a more reproducible and quantitative method for removing "outlier" standards. Thanks for your thoughts!
Having outliers on a regular basis is not good. It clearly reduces the trust in all results. The first and most important step should be to work on the robustness of your assay.
Given you can live with limited reliance on your results, you could use a robust regression technique that are not sensitive for outliers.
If your standard curve has enough data values and outlieres are rare, they usually have very little impact. You may check for points with a Cook's distance (cd) larger than 1. Remove the Point with the largest cd among these points, and refit the model. Repeat this until no points with cd > 1 are identified. If this happens to be more than, say 1-5% of your values, ring an alarm bell, as there is something wrong with your assay.
If possible, you might also run duplicates of either or/ both known standards or unknown samples. Or at least an occasional duplicate. This will give you some sense of the precision of the analytical procedure.
Sal Mangiafico , actually, all the measurements of the calibration curve are replicates. A large residual indicates that this measurement is suspicious, and the variance of the residuals gives an idea about the precision of the assay (additionally there are aspects like linearity and dynamic range to be considered, for sure). When measuring unknown samples, it is very common to measure triplicates. Many people do this to (more or less blindly) eliminate the one that is a bit far away from the two others (when there is such a case). However, except for very extreme cases this is not a sound procedure, as you know well.
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude