Critical comparison of colocalization algorithms on large datasets

Introduction: Great efforts have focused on identifying causal genetic variants shared between quantitative traits (QTs) and diseases. Several algorithms integrate association data and provide insights on the role of QTs in disease predisposition, such as Mendelian randomization, coincident associations and colocalization, all of them having benefits and drawbacks. Among these three, the colocalization aims at identifying shared haplotypes influencing both QTs and disease to find common molecular mechanisms. Here, we carry out a comparison of three colocalization algorithms (coloc, eCAVIAR and gwas-pw) reviewing their strengths and weaknesses. Materials and Methods: We tested three colocalization software using GWAS summary statistics of 272 immune QTs measured in up to 2,870 individuals from SardiNIA cohort and public data GWAS from a case-control metaanalysis on rheumatoid arthritis (RA) (>100,000 individuals). The three tools use different approaches based on Bayes factors to evaluate colocalization signals. Results: We considered as true positive the 270 traitlocus associations colocalizing (YC) or not-colocalizing (NC) with RA for at least 2 of the 3 tools, resulting in a total of 14 YC and 256 NC. Coloc and gwas-pw showed the highest concordance (100%), followed by eCAVIAR (98%). Conclusions: The comparison showed a high concordance among the colocalization algorithms. This suggests that the choice of tool does not have a strong impact on the colocalization results and the identification of links between QTs or diseases. We can assert that the colocalization is a robust approach to investigate connections between QTs and to advance understanding of biological mechanisms behind disease predisposition.