FeaturePlot of the first four featuresI am using FindMarkers() between 2 groups of cells, my results are listed but i'm having hard time in choosing the right markers. Do I choose according to both the p-values or just one of them? If one of them is good enough, which one should I prefer? I am interested in the marker-genes that are differentiating the groups, so what are the parameters i should look for?

             p_val avg_logFC pct.1 pct.2  p_val_adj
UBD   3.080608e-06 1.7755753     1 0.000 0.07683959
CFB   1.262305e-05 1.3989233     1 0.067 0.31485675
RGS13 1.548593e-05 0.9009480     1 0.200 0.38626551
PYDC1 1.896309e-05 0.9622537     1 0.133 0.47299636
BIRC3 4.099414e-05 1.0129472     1 1.000 1.00000000
ICAM1 4.379895e-05 0.8219610     1 0.200 1.00000000

as you can see, p-value seems significant, however the adjusted p-value is not. and when i performed the test i got this warning In wilcox.test.default(x = c(BC03LN_05 = 0.249819542916203, ... : cannot compute exact p-value with ties I am completely new to this field, and more importantly to mathematics. Please help me understand in an easy way.

  • $\begingroup$ You need to look at adjusted p values only. I'm a little surprised that the difference is not significant when that gene is expressed in 100% vs 0%, but if everything is right, you should trust the math that the difference is not statically significant. Why do you have so few cells with so many reads? Is this really single cell data? $\endgroup$
    – swbarnes2
    Commented Jul 30, 2020 at 3:24

1 Answer 1


The p-values are not very very significant, so the adj. p-value. You need to plot the gene counts and see why it is the case. It could be because they are captured/expressed only in very very few cells. VlnPlot or FeaturePlot functions should help. You haven't shown the TSNE/UMAP plots of the two clusters, so its hard to comment more.

  • $\begingroup$ I am working with 25 cells only, is that why? so without the adj p-value significance, the results aren't conclusive? $\endgroup$ Commented Jul 30, 2020 at 12:19
  • $\begingroup$ That is the purpose of statistical tests right ? At least if you plot the boxplots and show that there is a "suggestive" difference between cell-types but did not reach adj p-value thresholds, it might be still OK depending on the reviewers. But with out adj. p-values being significant and without seeing the data, I would assume its just noise. $\endgroup$
    – geek_y
    Commented Jul 30, 2020 at 12:47
  • $\begingroup$ I've added the featureplot in here. Is that enough to convince the readers? $\endgroup$ Commented Jul 30, 2020 at 14:02
  • $\begingroup$ For me its convincing, just that you don't have statistical power. Did you use wilcox test ? May be you could try something that is based on linear regression ? though you have very few data points. $\endgroup$
    – geek_y
    Commented Jul 30, 2020 at 14:46
  • $\begingroup$ yes i used the wilcox test.. anything else i should look into? $\endgroup$ Commented Jul 30, 2020 at 16:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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