how to convert beta effect size of SNP to a statistical value to be compared across tissues

Question

[rheab1230]

Hello everyone,

I want to use beta effect size of SNP across tissue for a comparison to see the shared ones. I want to use beta effect size, let say SNP1 and SNP2 are shared across 2 tissues and have similar beta effect size. I want to see how significantly are they similar or different?

Is there any way to convert beta effects for SNPs and convert them to pvalue or z score for comparison

Thank you

Answer 1

[LChart]

If the expression values were normalized, then the effect sizes (beta values) should be amenable to direct comparison. A beta value of +0.25 would indicate that heterozygotes (0/1) on average have an expression that is 0.25 Z-scores higher than homozygous minor (0/0).

[rheab1230]

Hello, So in my dataset during modeling for geneA expression using genotype in tissue 1 and tissue2. IN geneA model for tissue1: we have SNP1,SNP2,SNP3,SNP4 whereas geneA model for tissue2 we have SNP1,SNP2,SNP4,SNP5,SNP7. let say SNP1 is correlated with SNP5 so in tissue1: we only have contribution of SNP1 and have beta effect of 0.75

whereas in tissue2: we have correlated SNP1 and SNP5, so their individual beta effect is reduced now its 0.25,0.25 So I want to compare whether SNP5 which is present only in tissue2 geneA model, is it significantly different or specific?

My data are not normalized. I can try that.

[LChart]

If you're comparing effects, you really do need to keep the model specification the same, so the approach you laid out is wrong. You need to run expression by SNP[1,2,3,4,5,7] in both tissues; then you can compare the confidence intervals for SNP5.

[rheab1230]

Yes, I will do this approach and compare. But is there any way to compare beta effect size across tissues by converting them to z score?

[LChart]

So if you have your top table:

Gene        beta AveExpr     t       P.Value    adj.P.Val
Gene1       2.5      10.2        5.8     0.0002     0.002
Gene2       -1.2     8.7         -3.1    0.012      0.05
Gene3       0.8      7.3         2.2     0.035      0.1
Gene4       -2.0     6.5         -4.5    0.0001     0.001
Gene5       1.1      9.8         3.9     0.001      0.01
Gene6       -0.6     11.4        -2.0    0.053      0.2

you could do table$beta.z <- scale(table$beta) to normalize the beta values "relative to" the distribution of beta values within that tissue (or, you can do table$beta.z <- scale(ifelse(table$adj.P.Val > 0.1, NA, table$beta)) to restrict only to significant ones. You can then compare within-tissue normalized beta values across tissues.

This is really a strange thing to do -- not the least of which is because differences in sample sizes will basically determine the variability in the beta estimates, and therefore, the Z-scores. I'd strongly recommend re-considering your approach to your question.