Hi,
I'd like to compare between mass spec. results of 2 samples that I have (each sample has 3 biological replicates).
Following several filtration steps- I have a list of proteins that were identified in 1 sample (in all of the 3 replicates), but were not identified in the second sample... (and vice versa).
How should I consider these proteins in my statistical analysis? Can I consider them significant by default? Should I include them in one list with the other proteins that were (differentialy) identified in both samples and have values to compare?
Thanks!
If you're planning on doing quantitation you could add an arbitrary fraction of a spectral count to the 0 values and add this to all your samples and perform NSAF (which answers your other question about emPAI vs NSAF). This allows you then to perform stats, log-transform data etc. For simplicity, let's say Sample 1 Protein X has 5 (Rep1), 10 (Rep2), 8 (Rep 3) peptides and Sample 2 had 0, 0, 0, when you're working out protein fold-change from Sample 1/Sample 2, you can't divide by 0. By adding the spectral count (let's say 0.2) it is now Sample 1 (5.2, 10.2, 8.2 ) vs Sample 2 (0.2, 0.2, 0.2). If you're using NSAF, you also normalise by the length of the protein (bigger proteins = more peptides) and also the number of peptide/spectral counts in your dataset (variability in peptide #s in each run).
Depending on the biological system you're working with and the question you're trying to ask, changes to differential expression of proteins is more likely than the absence and presence of a protein in one sample to another. Caution this depends on your sample, an e.g. could be that you're artificially expressing a protein in a mammalian cell and you will be see 'absence or presence' of protein, or that you've treated one sample with stress and expressed a stress-specific protein/enzyme.
@ Albert Lee thanks!!!
maybe a stupid question, but is it important to add the value before normalization, or that I can just add a minimal value as NSAF?
Hi Chen Lesnik ,
I do agree with Albert Lee on the calculation part. Regarding the other query, if you have identified peptides corresponding to a protein in one sample and absent in other, in such situation, I myself would excuse myself in considering for my downstream analysis. The reason of absence in the sample 2 might be because the peptides might not be picked up due to low abundance or might not be there at all. So, one cannot claim saying its not present at all. But in such cases, one can go for more sensitive techniques validating such hypothesis. Thus, at the end, it all depends upon your goals and what you trying to achieve from your data.
Thanks!
Varsha
@ Varshasnata Mohanty Thanks! Would appreciate to have your advice whether it's important to add the value before normalization, or that I can just add a minimal value as NSAF?
Add arbitrary value and check using Shapiro-Wilk test if it fits a gaussian distribution. It's outlined quite nicely in the original paper by Zybailov 2006 JPR. Article Statistical Analysis of Membrane Proteome Expression Changes...
Dear Albert Lee , How to decide the arbitrary value to add, Whether to take 0.2 or 0.02? Is it based on the range of the values we have ? I am concerned because, when the values are in range 10-20, then addition of 0.2 will have less effect on the Fold change than compared to the data when the values ranges in 100-1000. Please let me know if I am wrong.
Santosh Kumar Behera To avoid dividing by 0 in the natural log transformation calculations, use the smallest value possible such that it does not change the total sum significantly. The value that you choose needs to provide the best fit to a gaussian (normal) distribution. All 0 spectral counts are replaced by a fraction of a spectral count (theoretically either 0.2 or 0.02 as you've designated), as long as the resulting ln(NSAF) dataset fits a gaussian distribution, which can be evaluated by the Shapiro−Wilk test. Whichever arbitrary value provides the 'better' distribution you would use that value.
- Badges
- Science topic