Hi Anders,

How big is the variation in quantified proteins within your samples and isolation method?

It is indeed possible (in my view) that you run into problems when having large variations in quantifiable proteins.For sure it does not allow any imputation of data for NaN. Almost all normalization algorithms rely on the concept that the majority of proteins will not change. So with any method you might have similar problems.

In my view the best would be to use the log2 iBAQ intensity for calculations. This was shown by Schwannhäuser et al that it correlates nicely with absolute protein content. In your MaxQuant analysis you need to specifically select iBAQ under Label free quantification (Global settings). Furthermore I would not use any normalization on these data.

Some general things:

If you want to use statistical test like a t-test that ask for independent samples, then you should pool the technical replicates before doing so otherwise your p values might be too low and of course wrong.

Regarding your questions:

I might not getting your biological question right ..  do you want to check which isolation method is the best e.g. number of quantified proteins or do you want to see difference on the protein level itself. E.g. protein x is higher in isolation method A compare to the rest.

In the case of protein numbers you would simply compare the total number of quantified proteins in each isolation method (I guess for that, calculating mean and standard deviation gives you enough information. (In this case you might want to keep the technical replicates separated to see how reproducible this result is. This tells you only something if you did the isolation twice and not simply measured the sample twice by LC-MS/MS.)

In case of comparing isolation methods on the protein level you would perform a anova analysis on log2 iBAQ intensities (Again: you should have already pooled your tech reps). You can do that in perseus using the p_n shortcut. Corresponding group would be that isolation methods are grouped together.  Note that an ANOVA does return (only) a p value. To find which sample(s) is / are changed you either calculated pairwise fold changes or perform Tukey's test (sometimes referred to as post hoc test).

More challenging might be the fact that you cannot calculate anything for a lot of proteins since it was only found in one isolation method. If you found a protein for example 3 times in one isolation method but never in the others it might be acceptable to assume that this protein is only present in one isolation method.

If you are able to do a statistical anaylsis on your dataset or better, is the quality good enough, I can not answer without having more details. In general, I would say  yes. 

Hope this helps

Hendrik

Hi Hendrik, thank you for the detailed answers!

My proteins varies quite a bit between the different methods. A quick rundown of the protein expression results looks like this (approximately):

• Isolation I: Identified proteins = 1500,Total intensity = 1,20e+10,
• Isolation II: Identified proteins = 1300,Total intensity =1,20e+10,
• Isolation III: Identified proteins = 800,Total intensity = 3,00e+10, :
• Isolation IV: Identified proteins = 600,Total intensity = 1,00e+10,

(I added Venn diagram to shows the unique proteins )

There are no major differences between the technical and biological replicates.

Questions

As to using IBAQ, I thought that IBAQ could only be used for protein quantification within samples (absolute quantification) and not between samples/methods (relative quantification). Am I completely mistaken here?

As to your question of what i want to look for, I want, if possible, to examine both the number of quantified proteins and differences in protein levels between methods. However,

Possible solution

Nevertheless, based on some further reading and your suggestions I've come up with a possible solution to the problem with accurate relative quantification.

What I'm planning to do is the following

(I would perform comparison between all methods, ie. method 1 vs 2, method 3 vs 1, method 2 vs 3, etc.)

• Use LFQ Intensity
• Filter out proteins that are not found in at least one of the methods being compared
• Log transform data
• Imputate for missing values
• T-test, etc.

My thinking is that doing this would make the assumption of similar protein populations between samples/methods furfilled.

Could this be a possible solution to my problem?

Hi, sorry for kinda late reply.

I think your suggested workflow could work. Be careful when imputating data. Always check the histogram of log2 LFQ intensities that you do not introduce too many data (e.g. for normal distribution). You would be able to see that by a kinda bimodal distribution. I think in your dataset this will be difficult when comparing different isolation methods because alone by comparing 4 and 1 you would have to imputate more values than you actually measured in method 4. Again, comparing between similiar isolation methods is fine I guess. :)

To compare different methods I really would simply compare the numbers of quantified proteins for your replicates.

For ibaq- indeed it was "developed" for aboslute quantification but it is also used for relative comparision in some studies and immunoprecipitations.

Best

Hendrik