Nonetheless, to ensure that the observed increases in de novo burden were not due to additional batch effects, we also performed a Poisson regression (Figure 4) to control for other factors potentially influencing de novo variant rate and detection. In iterative univariate multiple regression analyses, we observed that paternal age, sequencing coverage (percent of exome at 2× coverage), sequencing coverage uniformity (fold 80 base penalty), heterozygous SNP quality, and the number of de novo synonymous variants provided the best model for de novo coding variants. We used the size of the callable coding exome as an offset (Table 1; Table S1; Figure S3). The correlation between paternal age and de novo variant rate has been previously observed (e.g., Iossifov et al., 2012, 2014; Kong et al., 2012a; Neale et al., 2012; O’Roak et al., 2012; Sanders et al., 2012). Sex was not a significant predictor (Table S1). After controlling for these additional covariates in the Poisson multiple regression, de novo LGD variants still remained significantly associated with TD risk (Figure 4; RR 2.20, 95% CI 1.19–4.08, p = 0.01; RR
