The breakpoints of the commonly deleted region were estimated using a hidden Markov model. We calculated the Pearson's correlation coefficient between the inferred copy number genotype and normalized depth at base . If the base belongs to the commonly deleted region, then the true underlying correlation coefficient should be ; otherwise, it should be . It is apparent that the squared correlation coefficient multiplied by the sample size () asymptotically follows a distribution with a non-central parameter and one degree of freedom. Therefore, we fitted a hidden Markov model with the following two states: one corresponds to the null hypothesis of the test of Pearson's correlation (), and the other corresponds to the alternative hypothesis (). We used a standard Baum-Welch algorithm [32] to estimate and determined the maximum likelihood breakpoints located at 41,349,714 (95% CI: 41,349,709–41,349,715) and 41,381,486 (95% CI: 41,381,478–41,381,488) (Figure S10).
