Stephen, I disagree to skip technical replicates. It happens (too frequently from what I see in our labs) that one of the technical replictates shows a considerable shift in the ct (or Cq, how you write) values. Without having replicates, this would not even be identified as a possible problem. You are right that if the alternative of measureing 5 samples 3 times each was to measure 15 samples once, a single "outlier" or wrong ct would not be a big problem. But in practice, there are just, say, these 5 samples available, and one wrong ct among the 5 (single) measurements can cause considerable troubles.
And I still find it kind of funny to ask for the number of replicates a journal would accept. This is not good scientific practice. One should think about the precision of the estimates one aims/needs to get and decide based on this. This is a matter of the desired precision, not a matter of the gut feelings and habits of reviewers.
And if this was clear to Marianna, I am sure that she would have simply compared the achieved accuracy (with the 3 biological replictaes she has) and competently judged if this was acceptable for the porpose she had, and she would have been able to show this and to convince sceptic reviewers. A question like "what is typically enough" would never ever come into mind. At least not from a scientific prespective.
I am not blaming Marianna. This is all fine. It is a learning process. She might not have thought this way before and she might not have adequate support from her teachers and colleagues. I don't know. I won't render a judgent, neither about the person, nor the intentions, nor the skills. But I'd like to provoke with my answer. Not sure if this was successful.... but never mind.
Thank you for taking the time to answer my question. Sorry, I was not clear enough. My question is: let's hypothesize I want to publish a paper that includes RT-PCR results of different group samples for the same target/probe. I run each sample /target combo 3 times on each plate. How many times do I need to repeat a plate taking into consideration each different group is represented by 5-6 different samples in order the results be accepted for publication?
Honestly, Marianna, there is no common valid answer to su a question. It mainly depends on the reviewers of your manuscript. It could happen that a single measurement is accepted, others might moan that 3 repetitions are still too few. So do whatever is feasable and give it a try.
Apart from this very pragmatic answer there is actually a statistical problem behind the question. Replicates are typically performed to estimate a "reliability" of some estimate or conclusion. Here, technical terms like "variance", "effect size", "sampling distribution", "standard error", "confidence interval", "precision" and "power" come into play. I suspect that this all is far above the limit of your understanding of the problem. It might take some hours of discussion with your trusted statistician...
Jochen, thanks again but let's not make assumptions about somebody's understanding of statistical meanings on non existing evidence.
Really, my question is simple: based on experience how many and what kind of repeats are usually accepted for publication?
If you already have 5 to 6 samples (biological replicates) and 3 technical replicates per sample, this might already be enough. You might go for 3 plate replicates from which you can then calculate the technical error and per plate variation.
I have seen papers published with as few as two biological and three technical replicates...
Thank you Christian, the least described part of all M&Ms in so many papers is what is a repeat in PCR analysis.. We have 5-6 biological replicates and run triplicates of each on every plate. The question was how many times to replicate each plate under these conditions. Thanks again.
You should aim to quantitate RNA from at least two biological replicates. With regards to technical replicates, you may want to show the reviewers that your pipetting is up to scratch by doing triplicates for your standard curves, but there is then no need to do any more technical replicates, since they will usually be
Hi Stephen, thank you for all the suggestions.Good to know about the technical replicates. Diferences of Cqs because of pipetting errors have never been a problem for us so far. And, we routinely have groups of n > 3, run validation experiments for every new sample and/or new target combination. and run NTC and NRT on every plate. The MIQE paper is not very clear on this issue and many papers do not define "triplicates".
Stephen, I disagree to skip technical replicates. It happens (too frequently from what I see in our labs) that one of the technical replictates shows a considerable shift in the ct (or Cq, how you write) values. Without having replicates, this would not even be identified as a possible problem. You are right that if the alternative of measureing 5 samples 3 times each was to measure 15 samples once, a single "outlier" or wrong ct would not be a big problem. But in practice, there are just, say, these 5 samples available, and one wrong ct among the 5 (single) measurements can cause considerable troubles.
And I still find it kind of funny to ask for the number of replicates a journal would accept. This is not good scientific practice. One should think about the precision of the estimates one aims/needs to get and decide based on this. This is a matter of the desired precision, not a matter of the gut feelings and habits of reviewers.
And if this was clear to Marianna, I am sure that she would have simply compared the achieved accuracy (with the 3 biological replictaes she has) and competently judged if this was acceptable for the porpose she had, and she would have been able to show this and to convince sceptic reviewers. A question like "what is typically enough" would never ever come into mind. At least not from a scientific prespective.
I am not blaming Marianna. This is all fine. It is a learning process. She might not have thought this way before and she might not have adequate support from her teachers and colleagues. I don't know. I won't render a judgent, neither about the person, nor the intentions, nor the skills. But I'd like to provoke with my answer. Not sure if this was successful.... but never mind.
I would also not recommend to go for the least number of replicates possible. Two biological replicates are definitely not enough. It is beyond me how this can pass the reviewers.
But 5-6 is a good value to start. With a total of nine technical replicates spaced over three plates, the technical variance should be well taken care of.
The real challange: How to correctly determine the variance from all this data points...
Thanks for all the responses.
Christian, my concrete question then is what value do you use in the statistical analysis? If you do three plate replicates (all based on the same inter-run calibrator) per biological sample, do you take an average of the 9 values and use that one value in the analysis?
Basically, in this case you can use the average of all nine samples. In general, I would take the average of the technical replicates within a plate and then calculate the average of these averages over all three plates. Repeat for the average over all biological replicates. That way you can compensate for the loss/addition of replicates on each level and determine the variance for each replicate "level". This allows to identify problems more easily.
The real "trick" is the proper error propagation for the different replicate levels. One could argue that the variance of the biological samples is the only thing of interest and thus use only the average (of averages) of each biological replicate for the calculation of the variance, ignoring the variances of the technical replicates.
Better would be to use the rules for propagation of uncertainty for each replicate level, assuming that all errors are independent from each other (which is wrong but only results in an overestimation of the error). It would be nice to have the covariances for the intra- and inter-plate technical errors (as those are most likely coupled).
In conclusion, Jochens advice to get advice from a statistician is a good one (I am not really firm in the procedures and the terminology, especially in English). The only drawback in my experience is that these guys ask for so many additional measurements to get proper estimates for every parameter that the whole thing becomes really tedious and expensive.
For a biological interpretation only biologcal replicates are interesting. The observed variability is always a mix of the biological and the technical variance. But the sources of this variability are not important for the interpretation. The only problem is that the variance might be too large to draw any reliable conclusions. The biological variance is a fact, we can't influence it. The technical variance can be influenced by the choice of the method and by the number of technical replicates. When the technical variance is large compared to the biological variance, it is adviseable to use a more precise method and/or use averages from technical replicates. A technical replicate is every measurement that is taken from the same biological entity (indidivdual). No matter at what level a replication is performed (several wells in a plate, several plates, several cell cutures grown from the same primary isolate, ...). It might be methodologically interesting at what level the main part of the technical variance is introduced, e.g. to optimize the method or to compare different methods, but this is rarely the research focus. Further, there are some kinds of "biological entities" where it is not clear what an individual is (e.g. bacteria cells or cell lines from stocks). Here the question should be: if someone else in another lab who will repeat your experiment: what can he expect (based on your results)? In other words: at what level do I wish to generalize my findings? For instance, if I have a commercially available cell line what is available to any other lab, too, I don't need to bother with biological variance. If it is known that different lots of these cells can behave differently, then the "lot" would be the top-level entitiy to be replicated, since another lab would have to by another (different) lot.
Once your hands and reagents are all set, you start getting reproducible results in first practices, I will suggest minimum three times to be repeated, more is extra favor not needed.
.
- Badges
- Science method