As a complementary approach to testing previously described sequence motifs, we collated a large set of sequences likely to contain CNV breakpoints and used machine learning31 to discover new mutagenic motifs (Supplementary Methods and Supplementary Fig. 1.8). The motifs that we obtained, although significant, showed a modest enrichment for CNVs ranging from 1.2- to 1.5-fold. The most readily interpretable finding among these is a 14 bp CNV motif that is present in most Alu and SVA elements and has previously been shown to be associated with CNV breakpoints in Alu-Alu recombination events32 (Fig. 3b). This motif represents a binding site in the Alu secondary structure for the signal recognition particle ribonucleoprotein and is highly conserved across Alu elements.
