I have a large dataset of germline genetic mutations from a population of 1,000 individuals both affected and non-affected. I've already looked into individual mutations that are enriched in the affected population. Now I want to determine if the presence of two mutations in the same individual is more predictive of disease than having either one individually. Meaning does having both mutation A and B increase probability of being affected significantly more than having just A or just B. I want to identify a list of candidate pairings that show the greatest degree of interaction or predictive value.

The problem seems like something straight forward, but I haven't been able to find any literature dealing with this.

Thank you in advance!

  • 1
    $\begingroup$ In a linear model, this can be done with an interaction term. $\endgroup$ – user2974951 Jul 9 at 6:12
  • $\begingroup$ Do you have any references I can look at? I'm not quite sure if I know what that means $\endgroup$ – dannyrabiz Jul 9 at 12:09
  • $\begingroup$ You question is posed a little ambiguously, are you merely trying to find which variables to keep in a model because they influence the outcome? Do you think there may be interactions between the variables, such that two or more variables interact to improve predictions more than just including them on their own? As for finding pairs of variables, that's not really done. $\endgroup$ – user2974951 Jul 9 at 13:38
  • $\begingroup$ As for interactions, this is a standard topic in linear models, see en.wikipedia.org/wiki/Interaction_(statistics) $\endgroup$ – user2974951 Jul 9 at 13:39
  • $\begingroup$ I have a set over 200 million mutations. Since no mutation was significantly enriched in the affected population, I wanted to test if there is a pair of mutations that are enriched. P(Sick| A ∩ B) > P(Sick| A) and P(Sick| A ∩ B) > P(Sick| B) $\endgroup$ – dannyrabiz Jul 9 at 14:01

There is no model that will perform this as far as I know. The only option I see, if you really wanted to test pairs, is to build 3 models, one with A, one with B, and one with A and B. After this you check whether the last model performs better than the other ones. And you would have to perform this for every possible pair. If you have 200 million mutations (variables) this would result in a stupid amount of models and results.

You have to think of something else to do this, this is not really feasible.


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.