Beta-value for trans-eQTL

[broadinstitute]

Beta-value for trans-eQTL

[francois-a]

Great, thanks for confirming. I recommend filtering those out for trans-QTL mapping.

[francois-a]

Hi,
Thanks for pointing this out. This has been on the to-do list for a while. Added in 1ba37bb; the output (when return_sparse = True) now includes b and b_se.

[JonMarten]

Hi Francois, thanks for adding this.

I've noticed that as well as betas, the p-values have changed.

Using the same script each time, tensorqtl obtained through pip install tensorqtl, reports 8 SNPs as significant, but the version from pip install git+https://github.com/broadinstitute/tensorqtl.git#egg=tensorqtl gives no SNPs as significant. All other parameters and input data are identical. (It also runs a lot faster!)

Has the calculation of p-values changed?

[francois-a]

Hi,
Can you please provide specific data to reproduce the issues you're reporting?

The p-value calculation has not changed, and the output should be identical between versions. It's also unclear why you're seeing speed differences. I just re-tested these versions using the GEUVADIS example and am getting close to identical runtimes.

The only change in the latest commit is the addition of an imputation step for missing values at 1ba37bb#diff-c20eed07a12120ef8f33386bed4c6603R101. I added this for consistency with the cis-QTL code. This would change the p-values for variants with missing values, but such variants should generally be excluded from trans-QTL mapping (before running tensorQTL) due to the higher risk of being artifacts.

[JonMarten]

Thanks - I wonder if that could be it, I'll check. I have the usual data protection problems but if it's not being driven by missingness I'll try and recreate with toy or randomised data and share here.

As for the speed, if it's nothing you've changed, I strongly suspect it's just our cluster's quirky file system.

[JonMarten]

So I've checked by manually setting missing SNPs to different values and this imputation is definitely what's driving the effect I'm observing. Thanks!