logFC turned out to be "-lnf"

[bettycatherine]

Hi, I used SEURAT to identify conserved markers and it uses meta-analysis methods from the MetaDE R package, so I suppose this is a question about MetaDE. Some of the genes' logFC turned out to be “-Inf”. I looked up and cound not find a explaination for this. Would some one please tell me what this means and how to deal with it. Should I exclude these genes from my analysis?

Thank you very much!

Xue

[matianzhou]

Is it continuous data (and did you log transform the data?) or count data (are there concerns of excessive zeros)? How about the sample size? Those are some possible reasons that give inf logFC.

Best, Charles

Tianzhou (Charles) Ma, PhD Assistant Professor Department of Epidemiology and Biostatistics University of Maryland School of Public Health 2234M SPH Building #255 4200 Valley Drive College Park, MD 20742 Tel: 301-405-6421 Email: tma0929@umd.edu Website: (Department) https://sph.umd.edu/department/epib/bio/91666 (Personal) https://matianzhou.github.io/

On Wed, Feb 3, 2021 at 8:10 PM bettycatherine notifications@github.com wrote:

Hi, I used SEURAT to identify conserved markers and it uses meta-analysis methods from the MetaDE R package, so I suppose this is a question about MetaDE. Some of the genes' logFC turned out to be “-Inf”. I looked up and cound not find a explaination for this. Would some one please tell me what this means and how to deal with it. Should I exclude these genes from my analysis?

[image: image] https://user-images.githubusercontent.com/22362094/106830172-81aca600-66c8-11eb-92b0-03ced554e467.png [image: image] https://user-images.githubusercontent.com/22362094/106830213-9426df80-66c8-11eb-9523-ecd3b8f6e4f2.png

Thank you very much!

Xue

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/metaOmics/MetaDE/issues/14, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADZ3URD2NL6UHWVTEHRM3XTS5HXZVANCNFSM4XB4ENPA .

[bettycatherine]

Thank you, Charles, for your quick reply. My data should be count data, and I used smaller data, (subset this cluster), and there was no -Inf. So you are suggesting to use log-trasnformed data ? Thank you again!

best, Xue

[matianzhou]

No, you can definitely use the count data, but you need to make sure you choose the right method, e.g. DESeq2, edgeR or limmaVoom methods (where we call their packages). There is a chance that their tool (e.g. DESeq2) did some shrinkage on log2FC for count data and turn to some Inf values when there is excessive zeros (in all samples or in samples of one group). In your case, there is also a chance that there is outlier values in your full data but in your subset.

Best, Charles

Tianzhou (Charles) Ma, PhD Assistant Professor Department of Epidemiology and Biostatistics University of Maryland School of Public Health 2234M SPH Building #255 4200 Valley Drive College Park, MD 20742 Tel: 301-405-6421 Email: tma0929@umd.edu Website: (Department) https://sph.umd.edu/department/epib/bio/91666 (Personal) https://matianzhou.github.io/

On Thu, Feb 4, 2021 at 1:28 AM bettycatherine notifications@github.com wrote:

Thank you, Charles, for your quick reply. My data should be count data, and I used smaller data, (subset this cluster), and there was no -Inf. So you are suggesting to use log-trasnformed data ? Thank you again!

best, Xue

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/metaOmics/MetaDE/issues/14#issuecomment-773065714, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADZ3URCA6FPVWPN75OJSPQ3S5I5AHANCNFSM4XB4ENPA .