Omics data is super sensitive to experimental factors like date, technician, place, machine, kit, and something faint and subtle. So, you cannot assume any omics data reflect the true status. You should take them as distorted by non-linear systematic bias. You can compare (or analyze) samples if they are uniformly biased. We here say they are the "same type." However, you cannot directly compare samples if they are in "different types."
Please take a look at an example data set. The expression profiles are divisive in two groups, which do not correspond to biological parameters but replicate #1-5 and #6-10. It is impossible to know which group is of better quality, in other words, which group reflects the true status more accurately. We only know that this data set is a mixture of samples in "two types." This is an example of a microarray data set though, RNA-Seq data sets can neither be free from the same problem. We would say RNA-Seq more often see the problem, maybe because there is no physically universal device.
Having more samples get more chance of the risk. If the problem occurs, you will have to discard and redo a part of, or even all, operations. So, the estimated cost of a project is not only the cost of the first experiment (in case of no failure) but also the expected cost of re-experimenting. And the latter can be reduced by embedding risk hedge methods in the phase of experimental planning.
Additionally, following cases often cause problems in the data analysis and the risk hedge scheme is essential.
- The cost of re-experiment is high.
- Prospective studies.
- Ambitious experiments.
Please consult us at the phase of the experimental planning.