DivPop
This study used the MESA cohort individuals with genotype and expression data to generate cis-eQTLs and multi-ethnic predictors of gene expression and show that training set with ancestry similar to the test set is better at predicting gene expression in test populations, emphasizing the need for diverse population sampling in genomics.
PredictDB for use with PrediXcan and S-Predixcan.
MESA models are avaliable for download at- On the PredictDB page, click on
contributed/
and thenMESA-2018-05-v2/
to find the latest.db
files. These models are filtered byzscore_pval
< 0.05 andrho_avg
> 0.1, see here for explanation of performance statistics in the.db
tables. - Unfiltered models are in the
unfiltered_dbs/
directory above
.db
file with sqlite3
Example of how to pull model statistics from a $ sqlite3 AFA_imputed_10_peer_3_pcs_2.db
--view table schema--
sqlite> .schema
--output predictive performance R2 statistics to R2_values.txt--
sqlite> .headers on
sqlite> .mode tabs
sqlite> .output R2_values.txt
sqlite> select gene, genename, test_R2_avg from extra;
sqlite> .quit
here.
cis-eQTL summary statistics can be foundPLOS Genetics paper:
For more details, see ourGenetic architecture of gene expression traits across diverse populations
Lauren S Mogil, Angela Andaleon, Alexa Badalamenti, Scott P Dickinson, Xiuqing Guo, Jerome I Rotter, W. Craig Johnson, Hae Kyung Im, Yongmei Liu, Heather E Wheeler