Suppose I have a wild type and mutants A, B and AB. I take control (untreated) as well as treat each of them (i.e the WT and the mutants A,B and AB) with some treatment/s and perform RT-qPCR for 10-15 genes to see their differential regulation under the treatment. What kind of statistical analysis should be done in this case?
You can calculate standard error for your q-pcr data and plot on the graph. You can also calculate significance of differences between mean values of control and each treated sample using one way ANOVA and comparison among means can be carried out using Tukey–Kramer multiple comparisons test.
ANOVA if th groups are identical or non-parametrictest such as http://en.wikipedia.org/wiki/Mann%E2%80%93Whitney_U_test if they are not
What is the exact question you're trying to answer, and what is the exact experimental design (for instance, are treatments and control done on the same individuals or not) ? Until clearly stated, impossible to give sound advice for the methods to answer it...
I'd suggest to discuss this with local person (optimally with a statistician). This question won't be answered in a forum like this. You may get some hints, yes, but it seems that so many basics are missing, and finally you will only be confused or puzzled and possibly follow some advices you don't understand.
Apart from this, Emmanuel is absolutely correct: if someone should really help you, then you will have to provide ALL the details of the study and you have to be very clear and precise about the research question(s) (and the hypotheses) you have. Just to make one point (of many more possibe!): it is relevant if the mutations are known to act on the same path leading to gene regulation, or if they can act on parallel paths. This might even be your research question. If so, just for instance, something like a Bliss independence model may be indicated.
So do you think you will write all the details here, and that people will read this and spend a lot of time thinking about the design, the questions, the hypotheses and a possible appropriate analysis? And even then, will these people be able and willing to explain all this in a rather long correspondence in RG, so that you will understand the concepts and how to interpret the results? I wish you luck, but I find this unlikely. Maybe I am too pessimistic...
Hi Navjyoti,
As Emmanuel says it is difficult to give you an answer without knowing the design or what your question precisely is. I found this handbook very helpful for Biologists to get started with statistical analyses. I hope you will find it useful.
http://www.biostathandbook.com/analysissteps.html
and this guideline to choose the correct test
http://www.biostathandbook.com/testchoice.html
Best,
Thank you all for your suggestions. I have edited the question to make it more understandable.
Still not detailed enough. How many treatments? One or several doses? Are control and treated the same samples or not? If so, in which order? What is the question asked on the regulation: effect of each treatment, interaction between treatments, effect of some _family_ of treatments? Do you expect a down or up-regulation? How many genes (10-15 is not well defined)? and so on... What will you call a change in PCR biologically relevant for a regulation (power and sample size issues)? Not speaking of all technical details about performing the QT-rtPCR and the sample preparation that may influence the results, but that's another debate.
As said by many answers, without a precise, detailed experimental protocol, statistical analysis cannot be done (but anyway, an unplanned experiment is let's say difficult ot interpret anyway, if not useless).
Taking in account the information you provided, think about the following:
1º - take a look in the data distribution (normal vs non-normal). Probably, your data will not be normally distributed because gene expression data usually shows a great effect among treatments.
2º - Think about data transformation (e.g., log) for normality approximation.
3º - Use either ANOVA (normal data) or Kruskal-Wallis (non-normal data) with a post-test to perform the multiple comparisons.
PS: Using a statistical software (e.g., SAS, or R) you can perform all gene comparisons (i.e., all your 10 - 15 genes) at the same time. Also, you can correct possible confounders if you include the statements in the statistical model.
@ Peres Ramos Badial : just to be precise, normal distribution of the data is not the point to check. In fact, in case of a strong effect treatment, data seen as a whole cannot be normal, since they will be bimodal (or multimodal).
What is to check is (approximate) normality of the residuals after fitting a model on the data, ANOVA beeing a special case of such a model.
In simplest designs, this can be approached by testing normality of the data conditionnally to the treatment group, which I guess is what you meant. However, this is often less powerfull than residuals analysis.
- Badges
- Science topic