Norminal cis-windows for mQTL using TensorQTL

[hsun3163]

As noted in point 1 here: https://github.com/broadinstitute/tensorqtl/issues/68

there will be >400K CpG sites in the data compared to >20K genes. What would you suggest be a good cis_window for a CpG site, from a computational capacity perspective, for TensorQTL to analyze the data within reasonable CPU hours? 1Mbp radius may be too many tests?

Before this question got answered, this ticket will document some of the cis_windows other people use, although not necessarily using TensorQTL.

Cis_window is the number of bp up/downstream of the CpG site to determine what SNPs are to be included in the nominal association estimation.

[hsun3163]

One of the later (2020) mQTL paper used 50kb up/down of CpG site as the cis-windows for the nominal association pvalue. https://academic.oup.com/hmg/article/29/7/1154/5803132#203353336

This info will be organized into the analysis notebook were we actually use this window.

[hsun3163]

Another paper, https://www.nature.com/articles/s41598-018-35871-w#Sec7, used a way larger window.

Firstly the cis region to the methylation probe – defined as a window 2 Mbp each side of the target CpG site location – was investigated.

[hsun3163]

Another earlier paper(2016), https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4913906/#sec016,suggests using 500kb:

To distinguish between local (cis) and distant (trans) mQTLs a distance less or equal to 500 kb between a SNP and CpG site was used to define cis-mQTLs. All remaining SNP-CpG pairs were considered trans-mQTLs. In total we found 283,290,917,454 CpG–SNP pairs in the dataset, where 112,842,462 pairs were defined to be located in cis and 283,178,074,992 in trans.

[hsun3163]

@yuqimiao please post the windows u ultimately decided to use here.

[gaow]

whatever we do, it should not be smaller than the previous ROSMAP xQTL release ...