Statistics scare

If it doesn't fit into any other category post it here.
Proton Member
Proton Member
Posts: 1
Joined: Sun Jan 25, 2015 2:51 pm

Statistics scare

Postby bhattbh » Sun Jan 25, 2015 3:02 pm

I would like your opinion on the soundness of my experimental design and it's impact on the statistics of the data. So here's an elaborate description....
1) Four biological replicates of control and treated were made
2) The samples are processed independently and combined into ONE pooled sample just before 'zip-tipping' the final sample that will be loaded onto the column.
3) Each sample (control and treated) is run 5 times i.e there are 5 technical replicates/ chromatograms for each sample.
4) The data is then fed to Scaffold in such a manner that there is one biological rep supported by info from 5 tech. rep. i.e there are 5 raw files per category and only one bio rep per category.

I have a list of proteins that show great results when analysed so but the magnificent 4-fold, 9-fold changes are not supported by western blots for the same proteins- there is no change of any magnitude detected. The western blots are consistent over 12 replicates grown over a period of 6 months. What do you think is the problem?
Thank you for your response....I really need all the help I can get!

Angiotensin Member
Angiotensin Member
Posts: 42
Joined: Thu Dec 27, 2012 12:26 pm

Postby Christopher » Mon Jan 26, 2015 10:13 am

What type of quantification is this (e.g. SILAC, TMT, Label-free)?

What are the characteristics of the proteins that show the fold change values you observe? How many peptides? What is the variance in the quant values for proteins that are assigned more than one peptide within a replicate? How about across biological replicates?

Are you assigning any kind of p-value to your quantifications, or just relying purely on fold change values? For example, using something like Limma.

E. Coli Lysate Member
E. Coli Lysate Member
Posts: 107
Joined: Wed Dec 21, 2011 8:22 pm

Postby Infinity » Tue Jan 27, 2015 11:42 am

Did you normalize results somehow, either on the step of protein extract (use the same amount of protein from each condition) or when you obtained final FC ratios (so the mean of log2(FC) distribution is centered @ 0). Like Christopher mentioned look at the consistency in FC ratios obtained for different peptides corresponding to your protein of interest.
Obviously the best way would be to repeat the experiment, since you were able to do 5 technical replicates it is not a matter of sample amount (i.e. sample obtained from 1 biological replicate is enough for LC-MS analysis).

Angiotensin Member
Angiotensin Member
Posts: 31
Joined: Fri Sep 16, 2011 4:41 am

Postby gadsouza » Wed Jan 28, 2015 6:59 am

The sample you validated with WB: is that also pooled from 4 biological replicates from same "donors", or is individual sample without any pooling or a pool of different donors?

I personally do not like pooling samples. I can understand doing it to "save" instrument time or if the individual sample is at very low amounts. In your case you had pooled 4 biological replicates but made 5 technical replicates, I would had just injected each individual sample to also check biological variance.

For example, lets say you have Protein 1 that is 100x over-represented in just one of your controls, but is normally represented in the remaining 3 controls compared to treated samples. That protein is most probably just individual variation with no statistical significance. But when you pool your controls, you will "dilute" that one outlier, and when you compare control versus treated, it will look like that protein is 25x over-represented in all controls (rather than 100x in only one control).

If you then do a WB on a different pool of controls where all samples behave as the majority and there is no 100x outlier, your WB will tell you that there is no difference between control and treated.

Proton Member
Proton Member
Posts: 2
Joined: Tue Oct 15, 2013 10:27 am

Postby orsburn » Wed Mar 04, 2015 10:14 am

I think we'd all have a better idea of how to deal with the statistics if we had a little more information. Can you elaborate on the labels (as Christopher asked)? I'm a big fan of labeled quan, but there are different tools out there for different kinds of labels.

Return to “Other”

Who is online

Users browsing this forum: No registered users and 2 guests