Multivariate GWAS

For multivariate GWAS we used the MOSTest toolbox¹. MOSTest was developed in-house at NORMENT to improve statistical power of genetic analyses by leveraging the distributed nature of genetic effects across complex traits. In essence, MOSTest is a multivariate GWAS that allows for rapid and efficient analyses of multiple related phenotypes with large samples. The procedure is described in detail in Van der Meer et al.¹. In short, the procedure is similar to a regular (univariate) GWAS, but follows a few key steps. MOSTest first applies a rank-based inverse-normal transformation to create a normally distributed set of variables which is used for a univariate GWAS for each phenotype using a standard linear model similarly to the approach implemented in PLINK. Then, it estimates the correlation matrix between the genotype vectors for each univariate model through single random permutation. Next, it calculates a multivariate test statistic by drawing a comparison with the Mahalonobis norm where the probability of the observed test statistic is derived from a Chi-square distribution. Finally, it uses a gamma cumulative density function to draw a null distribution which enables calculation of the p-values beyond the 5e^-8 threshold. The output from MOSTest is a series of test statistics, both from univariate stream from the first step and the multivariate stream.

1.

van der Meer D, Frei O, Kaufmann T, Shadrin AA, Devor A, Smeland OB, et al. Understanding the genetic determinants of the brain with MOSTest. Nature Communications [Internet]. 2020 Jul 14;11(1):3512. Available from: https://doi.org/10.1038/s41467-020-17368-1