Background: Drug repurposing aims to identify new applications for already approved drugs. This approach is more cost-effective and reduces the risk of failure in human clinical trials, since the safety profiles of approved drugs are better understood. Though in-silico pipelines to identify drug repurposing candidates using GWAS data have been proposed, but the performance of such pipelines have not been systematically evaluated.
Methods: A disease gene expression signature can be imputed by integrating GWAS summary data with eQTL data from different tissue prediction models based on the GTEx-eQTL data using MetaXcan. This disease gene signature can then be used to query CMap to identify candidate drugs that ‘reverse’ the disease signature. Using a GWAS for low-density lipoprotein cholesterol (LDL-C) as ‘proof-of-principle’, we assess the sensitivity of this pipeline for identifying known cholesterol-modifying drugs.
Results: Our analysis revealed ‘liver’ as significant eQTL model for identifying disease gene signatures and a total of 530 significant genes (265 up and 265 down) were selected as different sets (5, 10….265) to identify negatively connected compounds using HepG2 cell line (liver cell line). Further enrichment scores revealed that ‘HMGCR inhibitors’ were enriched and significant (p-value < 0.05) based on mechanism of action for the set of 10, 20 and 25 genes. Furthermore, to validate our analysis, we have selected available relevant drugs (statins) to find overlap of negatively connected compounds. This analysis demonstrates that the sensitivity and specificity is very much dependent on the gene set size, eQTL tissue model used and CMap cell line.
Conclusion: A gene expression signature matching approach can be useful for identifying drug repurposing candidates but is highly sensitive to choice of tissue for imputing disease gene signatures from GWAS data and the choice of CMap cell-line used for generating drug signatures.