In recent years, genome-wide association studies (GWAS) have become one of the main methods to investigate disease aetiologies. GWAS leverage large-scale genomic data to identify significant associations between single nucleotide polymorphisms and complex traits such as diseases. This kind of genetic association is now used to inform drug development, with drugs supported by GWAS evidence being twice as likely to reach market than those without. One possible analytical approach to prioritise drug candidates from GWAS is called signature matching. In this framework, GWAS are used to perform a transcriptome-wide association study (TWAS) and generate a genetically predicted gene expression signature for a disease. This TWAS signature can then be compared to a perturbagen dataset such as the Broad’s Connectivity Map (CMap) and shortlist drug candidate based on similarities between the drug and TWAS signatures. While this approach has gained popularity recently, there is no consensus on the best analytical approach. The impact of the number of genes used to calculate TWAS-drug similarity, similarity metric, GWAS sample size, choice of tissue for TWAS, cell type for drug signatures or other key parameters have not been systematically tested to understand their impact on the final drug prioritisation. Here, using LDL-cholesterol, we systematically investigate how these different parameters impact the ability to prioritise HMGCR inhibitors using the GWAS-to-TWAS signature matching framework. We found that measuring drug perturbation in biologically relevant tissues such as the liver returned a strong negative enrichment of HMGCR inhibitors for LDL cholesterol (Normalised enrichment score±SD=-1.76±0.15) and a non-significant positive enrichment in HDL cholesterol (1.13±0.40), recapitulating known biological effects. Furthermore, we that other parameters such as number of genes used had a major effect on drug shortlisting. Together, we put forward recommendations on how best to perform a TWAS-based drug candidate prioritisation using the CMap dataset.