Fold change cut-off for SILAC/Dimethyl based quantifications

Search algorithms, post-searching processing, quantitation software, etc. Share and discuss software here.
karthikskamath
Ubiquitin Member
Ubiquitin Member
Posts: 52
Joined: Thu Dec 15, 2011 7:58 am

Fold change cut-off for SILAC/Dimethyl based quantifications

Postby karthikskamath » Wed Apr 01, 2015 9:06 pm

We did an experiments to check the experimental procedure bias on the distribution of the ratios. We took equal portions of same sample and labeled them with light and heavy dimethyl labels. We mixed them in equal proportions and performed LC-MS/MS on QE. We found following distribution of ratio (as shown in the figure)

Histogram.png


We observed few outliers in the ration which were outside the 2 times values of SD (which is 0.35). Following are our queries

1) In your experience what is SD and Median in such experiments?

2) We have observed two times SD as a cutoff in few papers for real/treated samples. How do you select the cut-off (up-regulation and down-regulation) values for diff expression in an SILAC/Dimethyl experiment? And why?
You do not have the required permissions to view the files attached to this post.

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

Postby Christopher » Thu Apr 02, 2015 7:25 am

1. The median and SD in the experiment you performed looks fairly normal. I would be more interested so see an MA plot of these data to help narrow down where the larger fold change values might be originating from.

2. You are correct that 2X SD is a common cut off. This really just illustrates the importance of replicates. If you were to use these two samples as true technical replicates, as they are, your 2X SD would be higher due to the variance of those outlier points. Typically, I will only use 2.5X SD as my ONLY cutoff if I have enough biological replication (at a bare minimum, 3 replicates) and my data set is too small for me to feel that I can do a good statistical test, such as from an IP-MS experiment. If you are working with whole-proteome expression values, such as you are here, there is no reason you shouldn't be trying to assign p-values using something like Limma. In this situation, I will typically use p-value<0.05 and a fold change >2.5X SD.

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

Postby Infinity » Thu Apr 02, 2015 10:48 am

I think that FC cutoff should never be used at all, nobody shower any correlation between FC and biological relevance. The only one thing that should determine up/downregualted peptides/proteins is a pValue cutoff (of course you have to correct for multiple hypothesis testing) and here you will get huge benefit by increasing the number of replicates. Even if log2(FC) is close to 0 but it is perfectly reproducible across all replicates - it should be considered as a good hit. Alternatively you might get log2(FC)=3 only in one replicate whereas in other experiments it will be close to 0 and obviously this should not be considered as biologically regulated. To summarize use correct statistics and base you findings on the solid pValue rather than arbitrary FC cutoff.


Return to “Bioinformatics”

Who is online

Users browsing this forum: Google [Bot] and 1 guest