Figure \ref{646437} shows the de novo CNVs in genes that are expressed in the brain (according to the Allen Brain Atlas). It makes no distinction between deletions and duplications. Something else that made a difference in the results was removing "well-know" CNVs. I used the same dataset the ADHD CNV papers used to discard well-know CNVs, under the idea that we're looking for rare variants. This cleaning process also excluded regions that (according to PennCNV) can be hard to infer signal, such as telomeres, centromeres, and others. Doing such cleaning wiped off the brain results, and significantly reduced the duplication CNV numbers in gene regions (compare Figures \ref{457923} and \ref{782746}).
That affected kid in family 10406 actually had lots of denovo deletions (Figure \ref{216248}), but it passed all QC metrics, so that's why I kept it. But if we zoom in the < 10 region of Figures \ref{533694} or \ref{216248}, we can see other families who also have disruptive (and protective) CNVs. So, the next question is whether the same genes are affected by those CNVs... nope. In other words, for example, even though families 10406, 10164, 10090, and 10173 all have kids with disrupting CNVs (Figure \ref{533694}), none of them code for the same gene. The 3 latter families have no intersections in gene lists, and although there are some intersections with 10406 (SP3, CREBRF; then NCAPG, LCORL, and ETNK1), I don't think it would be incredibly hard, as there are over 30 to choose from.