In microarray experiment on cell lines, should we take the technical replicate(loading sample from same passage number 3 times) or biological replicate (loading sample from cells of 3 different passages)
In my opinion, in any experiment both technical and biological replicates are equally important and they are 'not mutually exclusive'...while technical replicates will give and estimate to experimental variation ...the biological replicate will give and estimate of extent of variation in different samples....Furthermore...don't rely completely on microarray data...and validate the genes of interest by qPCR later.....here you can add more number of technical and biological replicates
if you are interested in a biological question, then you should do as many biological replicates as possible;
if you are interested in a technical question, then you should do as many technical replicates as possible (and here is would be even wise to not use any biological replication).
If the biology is your concern, then the biological replicates are the most important replicates. However, you may also include technical replicates, what can be beneficial when the technical variation is large compare to the biological variation (what is surely not the case for state-of-the-art microarrays). But if you have money for, say, 10 arrays and you could get 10 biological replicates, then I would not analyze only 5 samples on a replicate arrays. If you still have only 10 samples but you can afford 20 arrays, then replicate measurements can improve the results. However, the improvement will be small and likely not worth the money spent for the extra arrays.
Cell culture experiments come with their own problems regarding the replication units. If you use a cell line*, you actually have no real biological replication. What is relevant here is the level or layer (hierarchy) of the technical replication. The generation of the cell line might be seen as the highest level - this level is usually not achievable for replication (you have one stock in the fridge, possibly bought from ATCC or obtained from a colleague; this is it). At the next lower level you can take the preparation of a culture from this stock. The next level is the passaging and/or splitting of the cells, and eventually you come down to the level of the measurements (microarrays). How to deal with such things?
The simple answer is: do the replication at that level that gives the highest variability (this is often the highest possible level, because this will include all lower-level sources of variation).
However, this might not be optimal, from a statistical point of view. For instance, it might be better to use only few replicates of preparing the cultre but more replicates of plattings from these curtures. Or more measurements per plate, ... or the other way around. How to know what a good strategy is?
This could be tackled with an experiment including a sufficient number of replicated at each level. The data should then be analyzed with a mixed (hierarchical) model that will estimate all the variance-contributions at each level of replication. This can be used to plan subsequent experiments.
---
*if you are lucky your cells are primary isolates from animals or patients - in this case each animal/patient is a "biological replicate". The entire cell culture protocol from cell isolation over passaging to the measurement on a microarray is just the technical procedure required to produce a measurement from an individual (animal/patient).
In RNA-seq biological replicates are more useful than technical replicates to reach more coverage
Gierliński M. et al. Statistical models for RNA-seq data derived from a two-condition 48-replicate experiment //arXiv preprint arXiv:1505.00588. – 2015.
Liu Y., Zhou J., White K. P. RNA-seq differential expression studies: more sequence or more replication? //Bioinformatics. – 2014. – Т. 30. – №. 3. – С. 301-304.
good answers guys!! I agree with Jochen Wilhelm explanation. It depends on what you are doing. Do some technical replicate and see that there not too much variation. Biological replicate are more important for answering biological questions.