The identification of rare de novo mutations from CNV data on families is nontrivial. While CNV calls that show mendelian patterns of inheritance (which is the overwhelming majority of CNVs in the genome) are quite reliable, the fraction of CNV calls that are present in offspring and not in parents are enriched for technical errors, in particular false-positive calls (in the offspring) and false-negative calls (in parents). In addition, the enrichment of such errors is greater for smaller CNVs. To address these sources of error, we designed a set of algorithms for de novo CNV identification in families. The false-positive CNV call rate was controlled (maintained at 5%) for a wide range of CNV sizes by adaptive filtering of confidence scores (CSs), as described above. In order to minimize the number of false-negative calls in parents, the CNV region was directly genotyped using the MeZOD, and a confidence score was used to assign genotypes to the parents. Rare CNVs in children were called inherited if the CS was ≤ 0.04 in either of the two biological parents and de novo
