Although recurrent CNAs could be due to mutational hotspots we did not find a significant overlap between our recurrent CNA set and annotated chromosome fragile sites16 (17% overlap, P = 0.075). Recurrent CNAs could also arise if duplication or deletion of specific genes led to a selective advantage. To identify potential targets of selection, we defined peak regions of amplification (regions of maximum recurrence e.g. Fig. 2c and Extended Data Fig. 4) within each CNA and identified expressed genes (FPKM>0 in >10% of lines). Fourteen candidate regions contained fewer than six expressed genes including genes with establised roles in cancer progression (DOCK1, FATS, WWOX, STAG2 and XIAP)17–21. In regions with larger numbers of genes we searched for: (i) significant differential expression between lines with copy number 2 and 3 (ii) reported oncogenes from COSMIC20 and (iii) high scoring genes (top 2%) in a genome-wide siRNA screen for hESC cell proliferation22 (Fig. 2c, Extended Data Fig. 4; Supplementary Table 2). This approach identified BCL2L112 on chr20q11.21, EIF4A3, NOL11 and seven other genes on chr17q and UTP6 and SUZ12 on chr17q11.2. One
