You'll see in the X axis each family, and each circle is a child. Some families have more than one child, and there is always only one red circle (affected child) per family. Sometimes there is less than 2 circles per family, in which cases the kid was dropped during QC (or they're just overlayed in the picture). The Y axis counts the number of CNVs in each child, under different conditions (different figures). You'll also see some text under the figure, showing the families with biggest difference (in descending order) between affected and unaffected (red dot is higher, disrupted family), or between unaffected and affected (green dot higher, protected family). For example, in Fig \ref{782746} family 10090 has the biggest distance between red and green, and family 10131 has the biggest distance between green and red. The ideal result (i.e. lots of disrupting CNVs) would be lots of families with the red dot higher than green dots, and then when we looked at the (gene) location of those CNVs there would be some sort of consistency among families.
So, Figure \ref{782746} shows de novo burden by duplication CNVs, only in regions that code for genes. After going over hundreds of those plots (see things I tried below), these were the more interesting ones: