To achieve both precision and sensitivity, we combined four circRNAs prediction method, CIRI2, CIRCexplorer2, circRNA_finder, and find_circ to identify circRNAs. We obtained 935 cancer cell lines RNA-seq data across 22 cancer lineage from CCLE for use in the analyses. We required each circRNAs could be detected by at least two methods with backsplicing reads ≥ 2.
To examine the effect of genes on circRNA biogenesis, we assessed the correlation between gene expression levels with normalized total backsplicing reads. We defined the significant correlation between gene expression level and total backsplicing reads with absolute value of Spearman correlation > 0.3 and FDR < 0.05.
For each individual circRNA, we first classified cell lines into two groups based on the status of the circRNA expression pattern, then applied Wilcoxon test to identify drug sensitivity which is significantly associated with the circRNAs expression. We defined the significant association between drug sensitivity and circRNAs with FDR < 0.05.
For mRNA and protein level association, we first classified cell lines into two groups based on the status of the circRNA expression pattern, then applied Wilcoxon test to identify the significant association. For protein level and circRNA association, we considered FDR < 0.05 to be the statistically significant. For mRNA level and circRNA association, we considered fold > 1 and FDR < 0.05 to be the statistically significant. The exclusive or co-occurring association of circRNAs with mutation are identified by the algorithm DISCOVER. We considered FDR < 0.05 to be the statistically significant.