Questions regarding eqtl calls

Q:
I am trying to reproduce the eQTL calls published here with file name: Full_hg19_cis-eQTL. I’m having some difficulty reproducing the eQTL calls and in particular the P-values, and wanted to figure out where my pipeline isn’t matching.

1) I am unsure of the earth selection process on the super covariates sets. Currently, we try to reproduce the covariates selection using one hot matrix encoded covariates superset mentioned in the supplementary material (page 7) of this publication . We are curious on what covariates are selected (e.g.: brain bank covariates include multiple institutes, are all of them selected, or just some of them?).

2) We are unsure on which GTEx pipeline for EQTL calls were employed by the publication. We are currently using the GTEx pipeline mentioned here, but am wondering if the paper uses an older version of the GTEx pipeline that was previously available?

3) Another question is which datasets are fed into the eqtl calls? We are currently working with the capstone genotype datasets and TPM expression matrix published here with file name: DER-02_PEC_Gene_expression_matrix_TPM. We are wondering if the Genotype/Expression filtering were done directly on these files?

4) The last question is when we call eqtl using FastQTL, the nominal p-values (that have passed FDR < 0.05) are much larger compared to the p values your study published here with the file name: DER-08a_hg19_eQTL.significant (so it looks like we’re incredibly underpowered). I’ve attached a figure to illustrate the nominal p values reported in your files versus computed by us. We have used the Capstone genotypes and expression files (as described above), and though we should be somewhat underpowered relative to your study (because we are missing the GTEx genotypes/expression files, which need separate agreements), I’m not sure it accounts for the difference in p value magnitudes. I was wondering if you have any thoughts on which part of the pipelines we may have implemented incorrectly that could lead to such a huge difference?

A:
Here are some responses to your questions.

I am unsure of the earth selection process on the super covariates sets. Currently, we try to reproduce the covariates selection using one hot matrix encoded covariates superset mentioned in the supplementary material (page 7) of this publication . We are curious on what covariates are selected (e.g.: brain bank covariates include multiple institutes, are all of them selected, or just some of them?).
Here are the covariates we are using, you can also find the description in supplemental materials in our paper (http://papers.gersteinlab.org/papers/capstone4/index.html):

Top 3 genotyping principal components
Probabilistic Estimation of Expression Residuals (PEER) factors
Genotyping array platform
Gender
Disease status

We are unsure on which GTEx pipeline for EQTL calls were employed by the publication. We are currently using the GTEx pipeline mentioned here, but am wondering if the paper uses an older version of the GTEx pipeline that was previously available?
The detailed description of our eQTL pipeline could be found in Fig. S31 in our paper http://papers.gersteinlab.org/papers/capstone4/index.html.

Another question is which datasets are fed into the eqtl calls? We are currently working with the capstone genotype datasets and TPM expression matrix published here with file name: DER-02_PEC_Gene_expression_matrix_TPM. We are wondering if the Genotype/Expression filtering were done directly on these files?
You can find details in Fig. S31 in our paper http://papers.gersteinlab.org/papers/capstone4/index.html.

The last question is when we call eqtl using FastQTL, the nominal p-values (that have passed FDR < 0.05) are much larger compared to the p values your study published here with the file name: DER-08a_hg19_eQTL.significant (so it looks like we’re incredibly underpowered). I’ve attached a figure to illustrate the nominal p values reported in your files versus computed by us. We have used the Capstone genotypes and expression files (as described above), and though we should be somewhat underpowered relative to your study (because we are missing the GTEx genotypes/expression files, which need separate agreements), I’m not sure it accounts for the difference in p value magnitudes. I was wondering if you have any thoughts on which part of the pipelines we may have implemented incorrectly that could lead to such a huge difference?
I am not sure which genotype file you are using. But we cannot share the merged genotype file since we integrated some GTEx samples in the file. We are also using different covariates. So your results will be different from ours if the genotype, phenotype and covariates inputs are not the same.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s