Should you invalidate wells when performing a qualification of your liquid handling process?
Should you invalidate wells when performing a qualification of your liquid handling process?
You are conducting qualification campaigns on your liquid handling process (e.g. replication or reformatting) in your laboratory. Keep doing it, this is a good practice!
Qualifying a liquid handling process means to make sure the dispensing operations (occurring in the process) are conducted within an acceptance range or below a threshold criterion defined by the scientist. Most of the time, average volume (or absorbance, or fluorescence) and coefficient of variation (CV) are used as criteria to validate a dataset, thus a dispensing operation.
Yet when it comes to validate your dispensing operations by looking at the average volume and the CV, you are not sure if invalidating wells is a good practice or not.
Of course, You know there would be a good chance to improve the accuracy and/or precision by eliminating outliers in your dataset. However, you don’t want to skew the statistics so what could be acceptable and what might not be?
Well invalidation? No way!
Data are what they are, and you don’t want to bias the qualification of your liquid handling process.
Invalidating wells when you don’t know the reason may lead to keep a liquid handler active even though discrepancies in dispensing operations are noticed. Qualifying a liquid handling process also means identifying bad dispensing behaviors, which cannot be done if you systematically invalidate outliers without knowing the reason.
Indeed, invalidate a well because it “looks” like an outlier is definitely not recommended.
Well invalidation? Yes, but be careful and smart!
In order to select potential outliers, we suggest you choose a statistical parameter such as the Z-Score, then define threshold values (e.g Z-score values between -2.5 and 2.5 are acceptable). We are talking about Z-Score (which represents the number of standard deviations between a value and the mean) but you can use other statistical parameters.
The image below (Figure 1) illustrates our sayings. In this use case, we have three 384-well plates filled with a target volume of 1uL per well. By using the rule mentioned above, we highlight the out of range wells on the current plate (plate #1) and we identified the out of range for the two other plates (plates #2 and #3).
Then, you can choose whether to invalidate or not the selected wells. We strongly suggest you have a good reason to invalidate wells (e.g. air bubble, disabled channel).
Figure 1 - Well selection for invalidation from a Z-Score distribution (Discngine Qualification)
In the end, YOU have the choice!
Finally, remember that invalidating wells is not mandatory at all. Besides, there is no standard for well invalidations, so it is up to you to choose to invalidate wells. Some people will say that you should work with your data no matter what, and not alter them. Others will say that if you identify an exceptional reason for a particular outlier, you can invalidate some wells.
At Discngine, we are fine with both opinions, but we believe well invalidations require extra care from the scientists and should respect ground statistical rules. That’s why we developed this feature in Discngine Qualification hoping to satisfy most scientists in the community.