We did 4 experiments to compare the amount of certain proteins in treated and untreated cells. Each experiment was done separately. Because of the high cost of experiment, we were able to perform only one pair (one treated and one untreated) sample for each experiment. We want to see which proteins are differentially expressed (minimum 1.5 fold up/down).

First approach: We have compared protein levels of all 4 treated (as a group) with the 4 untreated (as a second group). There is of course variability between all experiments, because of the nature of the cells. We have a list of the proteins that are differentially expressed as a result of the treatment, however this list is not very long.

My question is (second approach): Can we compare the proteins levels pairwise for each experiment (treated vs its respective control) and make 4 corresponding lists, and then compare these four list using a statistical tool and find which proteins are consistently up- or down-regulated? DO you think that these two different approach will generate different lists of the affected proteins?

  • 2
    $\begingroup$ Welcome to SE Biology. Improving this question will help to attract an answer. Currently the questions are difficult to discern as they are buried amongst lots of text. A brief description of how the measurements have been made will also help. Finally, there are lots of typographical errors that should be corrected. $\endgroup$ – Michael_A Nov 22 '16 at 22:03
  • $\begingroup$ Can you please describe the experiments or just mention how these are different? Are the 4 controls essentially the same sample? If not then your first approach would be incorrect. $\endgroup$ – WYSIWYG Dec 1 '16 at 7:02
  • $\begingroup$ From what I understood, and thus based my answer on, the four treated and the four untreated samples are the same, but each pair of treated and untreated samples was done at different occasions. $\endgroup$ – BioGeo Dec 1 '16 at 12:18

If I understand you did a treatment to some cells and compared them with non-treated ones. Instead of running the four experiments at the same time, you did one treated and one untreated at a time. Then you did proteomics for each sample. Is the treatment the same in all four experiments?

Edit after the further comments of the OP: So, since the four treated samples are the same and the four untreated also (same conditions except from the treatment and the treatment is the same), then the way to go is as what your collaborator did.

Identification and quantification of the detected proteins is one thing and every sample of the four are replicates. Comparison between the two conditions is another thing. The software he uses for ID can combine the same samples and already perform the statistical analysis, so the list you have now has higher credibility in terms of the proteins it includes and their levels for your cells when they are treated or not.

Using only one of your replicates, although it might give you different number of detected proteins or different amounts of each, has lower credibility, because it's only one sample out of four. In plain words, the presence or absence or the amount of a protein might be an artifact or insignificant.

What has to be clear is that the comparison between treated and untreated conditions is done after you have received the statistically correct list of detected proteins.

Thus, and in accordance to what I had said before the edit, if you take the list for each treated sample and compare them with any of the lists of the untreated ones, it will lead to conclusions that won't be as statistically significant as when you combine all treated together and all untreated together (as your colleague did). In plain words, your conclusions will have a higher chance to be wrong.

Every statistical analysis you do yourself for each sample should eventually lead to a similar consensus as your colleague got using the statistical analysis of his software.

As a sidenote, considering how many types of proteins are in a cell, now that you have a quite short list might not be a bad thing at all and you can proceed by:

  1. Concluding that the treatment had minor effect in the proteins that you expected it would affect (if they are not present in your lists as significant different)
  2. Trying to understand, identify and hypothesize on the role of the proteins that made it in your comparison threshold, as they have a much higher probability of being indeed different between the two conditions.

Splitting the samples in your analysis might have a point if the conditions of the experiment were not exactly the same or the treatment level is different etc. In that case you could split the samples accordingly, but that would definitely reduce your certainty level for your conclusions.

| improve this answer | |
  • $\begingroup$ Thank you for your reply. It was impossible to do 4 replicates at the same time. Technically impossible. Therefore, we did 4 different experiments where we treated cells and we had an untreated control. $\endgroup$ – Leoncillo Nov 23 '16 at 16:16
  • $\begingroup$ Usually, we did proteomics for each experiment separately. Well, because we wanted to know our results ASAP. However, last time, we gave our samples - 4 pairs (i.e., 4 separate experiments of treated vs. untreated) to our collaborator who performs proteomics for us. He is doing HPLC - QqOrbitrap analyses for us. $\endgroup$ – Leoncillo Nov 23 '16 at 16:17
  • $\begingroup$ He always did the comparison between pairs of treated vs, the corresponding untreated, but last time when we gave him our 4 sets (4 pairs) he did something different, without our consideration - he somehow compared these samples as a two groups (this his explication), and thus he cannot extract the data for separate pairs (individual experiments). $\endgroup$ – Leoncillo Nov 23 '16 at 16:18
  • $\begingroup$ He wrote: "Fold changes are computed with all replicates and compared statistically. Extracting the data for individual replicate would decrease the ID power. Also, I would have to reprocess all the data files." When we insisted that we would like to have the list of the identified proteins that are up- or down-regulated for each experiment, he said that this is too difficult because of the way how he set up the software for this analyses, and that will take several months. $\endgroup$ – Leoncillo Nov 23 '16 at 16:18
  • $\begingroup$ My concern is that the "cumulative" list of the proteins that we have now because of this unexpected way of the performed analysis (two groups vs. 4 sets of treated and untreated) is much shorter. I am wondering if we perform the 4 pairwise proteomic comparisons and then will apply statistical tools - can we expect a much longer list of the identified affected proteins or no? Thank you for sharing with me your expert opinion. $\endgroup$ – Leoncillo Nov 23 '16 at 16:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.