Phenotype normalization

Prior to GWAS analysis, all data was corrected to eliminate systemic differences related to technical variability from cell passage and date of experiment. First, we averaged data coming from technical replicate of the same biological replicate. Second, we noted that some of our phenotypes (circadian period length, phase and amplitude) varied slightly but globally with the passage number of cell lines as well as with the date of the experiment. In order to eliminate effect of these confounding parameters we residualized our phenotypes using backward multiple regression model (lme function in nlme R package): starting from all variables included into the model we step by step eliminated less significant ones until all variables were nominally significant (p<0.05). Passage number we treated as ordinary numerical variables while date of experiments – as factors (dummy variables). Because all biological replications are non-independent, that is they belong to the same cell line, we used the cell line ID as a grouping variable in the lme function. Third, averaging residualized data coming from each biological replication we obtained the final phenotype value for each cell line. These values were used for all downstream analyses.