I'm using an R package to run interaction between genes using genetic markers (SNPs - single nucleotide polymorphisms). The package allows functions for phenotype (dependent variable) continuous and binary. I ran analysis once using continuous phenotype and once using binary. The output is a P-value between genes interacting.

With continuous
Gene_1-Gene_2 0.0008

With binary
Gene_1-Gene_2 8E-08

Two columns in output above - gene pair and their respective p-value.

The continuous phenotype is generating using residuals with covariates. Package doesn't allow to adjust for covariates thus I'm running analysis both ways.

Is there a way I can compare results from these two steps? Say correlation or something?


1 Answer 1


P-values are comparable, but I am not sure what your research question is. Do you want to know if there are gene pairs that interact with respect to one phenotype but not with respect to another? If that is the case, try to apply Fisher's z transform to the p-values (so that they become normal distributed under the null hypothesis) and look at scatter plots. Such a scatter plots might identify three classes of gene pairs: - the majority are in a cloud centered in (0,0), meaning no interaction - some are positive outliers for both phenotypes (interacting) - some have clear interactions with respect to one phenotype (positive) but not with respect to the other (close to zero).

Can you get the package to distinguish between positive and negative interaction? That might be helpful.

You might find that genes that interact with repect to both tend to have smaller p-values with respect to one phenotype than the other. If the two data sets have different sample size, this will trivially be the case (large n -> small p), but could also be related to the accuracy wrt which the phenotypes are recorded. Or it could be that one phenotype has a stronger genetic component than the other.

  • 1
    $\begingroup$ Actually, the P-value from binary are way off than continuous, or vice-versa. The binary doesn't adjust for covariates, so that explains so high degree of difference in P-values. It would have been great if package could give me direction of interaction, but it doesn't. The two data sets are same. I'm using same input individuals, but the phenotype is changed from binary to continuous (adjusting for covars). $\endgroup$ Mar 19, 2019 at 1:13

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.