We also performed a Poisson regression to control for factors influencing de novo mutation rate and detection, such as paternal age and sequencing coverage (Iossifov et al., 2014; Kong et al., 2012b; O’Roak et al., 2012; Sanders et al., 2012), respectively. We used the Akaike information criterion (AIC), implemented in R, to assess the relative quality of different Poisson models for predicting the number of de novo coding variants. During model selection, we assessed potential covariates versus the response variable of coding de novo mutation rate in SSC control trios, and without including affected status as a covariate. We chose to look in the SSC trios only, because we observed that most batch effects observed across the cohorts were strongly correlated with phenotype status. However, repeating these steps across all of the cohorts resulted in the same final model (not shown). We determined that paternal age, sequencing coverage (percent of exome at 2× coverage), sequencing coverage uniformity (fold 80 base penalty), and heterozygous SNP quality provided the best model. Additionally, however, we reasoned that the number of de novo synonymous
