MatrixeQTL FDR and Q-Qplot

Question

[toyan.j.p]

What would be the cut off for FDR in matrixeqtl, is it 0.05 as the case with pvalue?.Is there any guidelines for it?I couldn't get the same from the matrix eQTL paper. Is it mandatory that you apply correction if its a targeted experiment involving a certain region of chromosome and a single phenotype?I would also like to know how to interpret Q-Q plot from MatrixeQTL?

Answer 1

[brentp]

The FDR is based on the number of tests in your experiment, so yes, you should still use FDR correction. 0.05 is a reasonable FDR cutoff.

The QQ plot should follow the 45 degree (y = x) line except at very low p-values (very far to the right). If your points are far above the line starting from higher p-values, then you have inflation and you need to resolve that. It is usually caused by not adjusting for an important covariate or by population stratification.

[Laura]

Hi Brentp,

I recently performed a MatrixeQTL analysis and was wondering about my QQ plot. Your above answer was already really useful, I just hoping to clarify something a bit more.

You mention that you would only expect the QQ plot to deviate from the (y=x) line at very low p-values. My p-values start to deviate a from the 45 degree line at p<0.001 (local) and p<0.0001 (distal). Would you say those P-values are sufficiently low for their deviation from x=y to be acceptable or should I investigate if I have inflation? I did incorporate covariates in the analysis