We combined both EUR and YRI data sets from

Paper: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3918453/
DataFtp: http://jungle.unige.ch/~lappalainen/geuvadis/

To generate the positive sets we started with the FDR5 data set, 
and accepted only autosomal SNP's associated with exon and gene 
expression levels. Specifically, we excluded from 
positive & control sets:
- trqtl
- non-SNV variants such as indels
- annotated CDS

The resultant positive set had 6,520 unique, noncoding, autosomal positions.

The control set for the paper was generated from all dbSNP v137 SNP's. 
We also verified against a second control set consisting of all of 
the SNP's tested as QTL candidates by Tuuli Lappalainen, 
(9,7943,317 autosomal SNV), less any variants that had 
a p<.05 correlation with transcription from an exon or 
gene. This removed all identified genic & exonic QTL's 
as well as some positions that showed correlation with 
expression, but not enough to pass a statistical test. 
This left 7,645,533 positions.

-- Brad