Hi Swati, 

Data collected from the same patient at different time points are likely to be correlated -  1-hr value being influenced by 0-hr value which in turn shall influence the next value.

Doing independent t-test will ignore this resulting in biased results. One has to do regression to account for the correlated values. The options are repeated measures ANOVA and generalised estimating equations (GEE) - population averaged model. 

I am familiar with GEE and can help you if you indicate the statistical program you use. 

Jeeva 

Dear Swati,

I assume you have done at least one (better 2) reference gene that was validated to be stable under your experimental conditions and are familiar with the MIQE guidelines...

There are a few options for statistical analysis. My personal favourite is REST2009 which you can find under the link below. I really like it because it does not require you to make any mathematical decisions or calculations yourself (although Jeeva is entirely correct in which tests are appropriate!). You simply put your Ct values in and it comes up with a properly done analysis and a nice graph too.

The other option is to use a program called GraphPad prism (free 30 day trial) and do the analysis yourself (repeated measures ANOVA), There are other stats packages of course, but I like GraphPad as it is targeted at biologists not mathematicians and helps you decide which tests are appropriate for any given experiment.

You could of course go for a full qPCR analysis program like GenEx or qBase which will also do the analysis for you (2nd link below).

Petra

http://rest-2009.gene-quantification.info/

http://www.gene-quantification.info/

Dear Swati

Taking your things into account, I too suggest you to go for REST2009 suggested by Petra. I am using it and its really good and you get a clean analysis.

Best wishes

Thanks Jeeva Sankar, Petra Disterer and Raj Bhaiya for your replies.

The three BRs that I am using have different plants, so none of the BRs are dependent. Same goes for the time points, all the time points have tisuues from different plants. Wouldn't this make all the time points independent of each other!! I am attaching two papers where they have used different time points and used T-test for analysis!

Please correct me if I am getting this wrong!

Swati

http://bmcgenomics.biomedcentral.com/articles/10.1186/1471-2164-15-1130

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4159813/#!po=22.7273

If the tissues are from different plants you should go for a regular ANOVA (no repeated measures) you should check before if your samples fit to a normal distribution because if they don't, you have to go for non parametric test (Kruskal-Wallis test). These multiple test show differences between all the points (not only 0 vs 1h, o vs 2h but also 1h vs 2 h etc), if you are only interested in the difference between the times and 0 you can go for t-test (if normal distribution) or Mann-Whitney (if not), be aware that in this case you are not showing an actual time course curve because you are not showing that 1h is different from 2h.

Thanks Roberto for your reply.  So if in case I compare 1h control vs 1h stressed samples (tissues are from different plants), I should use T-test?

If your control is not 1 h you can not use t-test, if you want to compare 1h vs 1 h stressed you should use 1h as control, if you are using 0 as control in order to compare 1h vs 1h stressed you should use either non repeated measures ANOVA or Kruskal Wallis, I hope this help

Dear Swati,

Like many biologists, statistics isn't my favourite thing either! And I completely understand how it is difficult to understand, so I'll try and put in in easy words. Any real stats (Roberto and Jeeva seem to know what they say), please correct me if I am wrong!

The t test really is a much misused test. It should only be used when comparing two groups with normal distributions (see below link for explanations of the stats terms that have been thrown around here!). "normal distribution" also known as "Gaussian" distribution has the form of a bell curve with 95% of values within 2 standard deviations of the mean (peak of the curve; see Mathisfun link below).

Many people and papers use the t test incorrectly (many reviewers do not understand it either!).

Simply doing multiple t tests to compare more than two groups is not the right way to do it!

More generally: Just because something is published doesn't make it right!

IMHO, as scientists, we should always strive to do the best possible science and not settle for doing things "as they've always been done" when they've always been done wrong.

Back to your issue: I assume you want to say something about how a particular factor changes in response to stress over time. If you simply wanted to look at 6 h versus 0 h, why would you take the intervening timepoints?

So you have the following groups: a number of plants before stress (0 h) and a number of plants at 1,2,3,4,5 and 6 h after the stress (this equals 7 groups in my count, so definitly more than 2; this means t test excluded). So Kruskal Wallis is the way to go or ANOVA if you expect the distributions of your samples to be normal. I would say that for experimental purposes the Kruskal Wallis should be your first stop before ANOVA - because low n numbers of samples are common in experiments, and with low n you cannot know if the distributions are similar!

Exclusions from that guideline are: flow cytometry data (where every cell = data point); digital PCR (again every well = data point) or similar assays where n >10-20.

I hope that you have at least 3 plants each (n=3), but a bigger number would be better. I would also recommend to do a control group for each time point (not a statistical neccessity, but really more an experimental meaningful control - after all how do you know that your factor of interest does not change in non-stressed plants over that time? There could be some circadian mechanisms involved or the control plants have been watered when the stressed ones have not...).

https://www.graphpad.com/guides/prism/7/statistics/

https://www.mathsisfun.com/data/standard-normal-distribution.html

You have time  time course data for quite a number of different time points. This actually screams for a regression model.

The options here are manifold. Firstly you should plot the response against the time and think of a sensible yet simple functional model that could describe the change in response over time. Is it a simple (linear or exponential) increase or decrease? It is "peaking" or "switching" (i.e. describing a sigmoidal curve). If you have identified what a sensible model might be then you can fit the coefficients of this model to your observed data, and then you can ask specific questions about the model parameters (e.g, is the peak hight different between the groups? is the reaction time different? is the (initial) speed of change different? etc.

If the model is not linear (in its coefficients), you have to go for a a non-linear regression.

Thanks a lot Petra for  your explanation. It is very true first understanding and then applying statistics is hard (atleast for me). Now, I have got an idea on which direction to move from here. Thanks for your gudiance :-)