I'm working on cancer mutation analysis from system perspective and ran across a statistical problem. My research subject is "network motif" in cancer. In my case, there are 53 different motifs and each of them has two kinds of mutations (i.e. pathogenic or neutral) with a normalized count.


enter image description here

Row names are motif IDs, third column is ratio of pathogenic/neutral.

My task is to determine a threshold of this ratio (e.g. 2). And I want to say that any motif whose ratio above this threshold indicates that the difference between pathogenic and neutral mutations on this motif is statistically significant.

I would appreciate help with choosing a statistical test for this problem.

  • $\begingroup$ I don't think you've presented sufficient information for us to judge whether it is even possible to define any sort of meaningful threshold. $\endgroup$ – mkt Aug 1 '17 at 22:27
  • $\begingroup$ Thanks for pointing out. My current research objective is to find out which "network motif" out of 53 in total has large ratio, which indicates this motif is preferable for pathogenic mutations but not neutral one. But I don't know how large this ratio is enough, so I need a threshold to determine significantly "large". $\endgroup$ – Haoran Chen Aug 1 '17 at 23:18

Your threshold value is going to be dependent on the variance of your samples. I would recommend one of the tests found here: T-test

You must calculate the critical value which will be the t value of the sample degrees of freedom and the desired confidence level found here: T-table. Then you calculate the sample t statistic based on the two samples and compare the critical value to the t-statistic.

  • $\begingroup$ Thank you so much! I just notice you answered me, sorry for the late reply.. Could you please explain a bit more of the second step about calculating the sample t statistic based on the two samples and compare the critical value to the t-statistic? I thought t-test is used for comparing the means of two samples. How can it determine the significant threshold of the ratio between two independent samples? Thanks in advance!! $\endgroup$ – Haoran Chen Aug 7 '17 at 14:03
  • $\begingroup$ @HaoranChen I'm not versed in your field of study but I'm assuming a motif is a change from some baseline? If so you'd compare each motif to the baseline data with a t-test. If not then I may not understand your question fully. $\endgroup$ – Acumen Simulator Aug 7 '17 at 15:54
  • $\begingroup$ Thank you so much for your prompt reply! I'm sorry that I haven't expressed the question clearly. We can regard 53 motif are 53 "patients", and there are two kinds of independent data attached to each patient: pre-treatment and post-treatment. We want to determine if a patient has a significant response to this treatment, if so we can apply further research on these well-responded patients. Therefore, we need to determine a threshold saying that if post-treatment/pre-treatment is above this threshold, we can say that this patient's response is significant. $\endgroup$ – Haoran Chen Aug 8 '17 at 16:24
  • $\begingroup$ @HaoranChen ok so the analysis is the same. You do however need a baseline to compare the motifs to.... say a motif 0 which would be the control group or placebo group where there was no treatment given. Then perform the t test on each motif vs motif 0. $\endgroup$ – Acumen Simulator Aug 8 '17 at 16:29
  • $\begingroup$ Thanks a lot! I think the control group for me is that there is no difference between post- and pre-, namely two identical data. I'm still a bit confused. So the null hypothesis for this problem should be: There is no difference between post- and pre-treatment? And then perform t-test on each motif based on this null hypothesis? $\endgroup$ – Haoran Chen Aug 8 '17 at 16:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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