Question: Best cutoff for fold change of microarray data
0
gravatar for Expert
8 months ago by
Expert30
Expert30 wrote:

Dear all I have 6 microarray data set and I want to select a cut off base on adj. p.value and fold change to find DEGs. About adj. p.value, I know a cut off less than .05 or .01 are the best but what is the best fold chang cutoff if we want to compare different datasets? all datasets are affymatrix and normalized base on R package.

next-gen • 595 views
ADD COMMENTlink modified 8 months ago • written 8 months ago by Expert30

All dataset are affymetix and normalized on GEO2R from NCBI. After that I want to select a cut off and I know it is different among veriety of data.

ADD REPLYlink written 8 months ago by Expert30

Thanks - please give your response to my answer (below). This helps to maintain the fluidity of the conversation.

I moved your answer to a comment.

ADD REPLYlink modified 8 months ago • written 8 months ago by Kevin Blighe41k

nfs.falsafi : Please do not delete posts when they have received a comment/answer.

ADD REPLYlink written 8 months ago by genomax65k
1
gravatar for Kevin Blighe
8 months ago by
Kevin Blighe41k
Guy's Hospital, London
Kevin Blighe41k wrote:

I assume that all of your datasets are the same array type / version? How have you processed and normalised them - all together?

There are no correct cut-offs. The cut-offs change in each and every study. So, not even P 0.05 or 0.01 should be regarded as the best cut-of for statistical significance. Also, you should be using FDR-adjusted P values, i.e., Q values.

For fold-changes, there is neither any correct cut-off. Some use absolute log (base 2) fold change 1.0, whilst others use 2.0. Others don't set any fold-change cut-offs and instead just rank the statistically-significant genes based on fold-change.

You are the analyst and you are in control.

Kevin

ADD COMMENTlink written 8 months ago by Kevin Blighe41k

Thanks Kevin, I got it. in my data, I think FDR<0.05 and fold change >1.5 and <-1.5 for up- and dw-regulated genes respectively are the best (based on Mark R Dalman et al paper) but after normalization with R packages, we have log transformation of fold changes and I know if log FC is 2 it means 4 fold change for gene expression. and logFC -1= FC 0.5. it means, we don't have negative measure for fold change in statistic but have in biology. how shell i put it! what is the similar value of FC -1.5 for log FC? (?logFC=FC-1.5)

Thanks in advance

ADD REPLYlink modified 8 months ago • written 8 months ago by Expert30
1

Fold change can not be negative, but log-transformed fold change can. Imagine your initial expression is 100 and after a treatment it decreased to 50. The fold change is 0.5, and log(2) fold change is -1.

ADD REPLYlink written 8 months ago by grant.hovhannisyan1.5k

Posted a response here but ignore it. This thread has received many comments today, some removed.

ADD REPLYlink modified 8 months ago • written 8 months ago by Kevin Blighe41k

nfs.falsafi, as per Grant's comment, above.

ADD REPLYlink written 8 months ago by Kevin Blighe41k
1
gravatar for grant.hovhannisyan
8 months ago by
grant.hovhannisyan1.5k wrote:

While there is no ideal cutoff for p-value (which has to be corrected for multiple testing) and fold change, you can have a look at a similar or reference studies and choose cutoffs based on them to be able to make fare comparisons with already published results.

ADD COMMENTlink modified 8 months ago • written 8 months ago by grant.hovhannisyan1.5k
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1196 users visited in the last hour