Hi everybody!
I wonder whether there exists some minimum threshold in xfold protein expression change for differential proteome analysis (e.g. effect of a certain treatment on the proteome of a certain tissue)? I know that in differential transcriptomics via microarrays and NGS a 2-fold change is generally accepted as threshold from which on a change in transcript abundance is considered to represent signficant regulation.
I appreciate your input!
Thanks and regards,
Daniel.
As one of the authors of the RNA paper that @Laurent highlighted I'd like to add a comment.
Although our study is on RNA-seq there are parallels with any quantitative 'omics experiment. There are several features that need to be identified before you can assess whether a change is a relevant in your study:
1) experimental design. Good experimental design will be clear on what is being measured and that the technique used is appropriate for doing so. It will also factor in robustness to avoid single points of failure ruining the whole experiment.
2) quality control of the data. Before any analysis is done ensure the data are as expected. If they aren't, find out why before blindly carrying on regardless.
3) variability. What is the variability in your system? This can only be identified with a suitable number of *biological* replicates (nb: 3 is rarely enough)
4) statistical significance. Once variability is assessed the statistical significance can be determined using a correct statistic and adjusting for multiple-hypothesis testing. This is critical for avoiding false positive and false negative results. This is pretty well understood in RNA-seq and microarray. Not so much in proteomics.
It's very important not to confound statistical significance with biological significance. Given enough replicates in a low-variability system you will find almost all genes/proteins as being significantly changing. A fixed fold-change threshold for all experiments is not appropriate, but likewise spending 6 months following up a statistically significant change which only shows a 5% difference in expression is probably a wasted effort.
So, in summary, to answer your question. No a 2-fold threshold is not an acceptable test for significance. It's too simplistic. You need to look at a combination of statistically significant changes together with their fold-change and effect size.
Hi Daniel,
I don't think there is, due to the fact that -in contrast to a lot of transcriptomics setups- proteomics setups differ quite a lot, for example with respect to sample size and method of quantitation.
Depending on the resolution and size of your study, I think it is advisable to not only check the fold change, but also the statistical significance of that particular difference. In some cases that can mean that a 1.5-fold change is 'real', while a 10-fold isn't.
What proteome are you studying and what analytical methods are you using?
Hope this helps!
Hi Eef,
thank you very much for your quick and helpfull response.
I see your point. My question does not refer to my own research but data published by another group. They used isobaric labeling prior to tandem MS to compare proteome adaption to a certain stimulus. They set their threshold at >1.2 fold change in peptide abundance (p
Hello Laurent,
thank you very much for your input!
This helps a lot!
Regards,
Daniel.
As one of the authors of the RNA paper that @Laurent highlighted I'd like to add a comment.
Although our study is on RNA-seq there are parallels with any quantitative 'omics experiment. There are several features that need to be identified before you can assess whether a change is a relevant in your study:
1) experimental design. Good experimental design will be clear on what is being measured and that the technique used is appropriate for doing so. It will also factor in robustness to avoid single points of failure ruining the whole experiment.
2) quality control of the data. Before any analysis is done ensure the data are as expected. If they aren't, find out why before blindly carrying on regardless.
3) variability. What is the variability in your system? This can only be identified with a suitable number of *biological* replicates (nb: 3 is rarely enough)
4) statistical significance. Once variability is assessed the statistical significance can be determined using a correct statistic and adjusting for multiple-hypothesis testing. This is critical for avoiding false positive and false negative results. This is pretty well understood in RNA-seq and microarray. Not so much in proteomics.
It's very important not to confound statistical significance with biological significance. Given enough replicates in a low-variability system you will find almost all genes/proteins as being significantly changing. A fixed fold-change threshold for all experiments is not appropriate, but likewise spending 6 months following up a statistically significant change which only shows a 5% difference in expression is probably a wasted effort.
So, in summary, to answer your question. No a 2-fold threshold is not an acceptable test for significance. It's too simplistic. You need to look at a combination of statistically significant changes together with their fold-change and effect size.
- Badges
- Science method