Fig. 2: Assessment of the IRAVNet approach using TCGA transcriptome and exome sequencing data.

a The flowchart showing the classification of genome level mutation status for IRAVs. Here, according to sequencing depths, the variant counts (the numbers of intron retention reads), and variant allele frequencies (VAF, the ratios of intron retention reads to the total number of sequencing reads covering the corresponding exon-intron boundaries), the status are classified into “germline,” “somatic,” “somatic or germline,” “ambiguous”, and “false positive.” b The number of IRAVs categorized by the inferred mutation status for the identified IRAVs determined by the above procedure. c The boxplot showing how the intron retention is specific to the samples with the IRAVs. The ends of the boxes indicate lower and upper quartiles; center line, median; whiskers, maximum and minimum values within 1.5 × IQR from the edges of the box, respectively. For each IRAV, the Z-value comparing the ratios of intron retention between samples with the IRAV and other samples in the same cancer type group is computed. We observed that most of the Z-values were above the reasonable threshold (>2), strongly suggesting that most IRAVs certainly generate intron retention. The information, such as the sample size for each box, is provided in the Source Data file. d Landscape of IRAVs in frequently altered cancer-related genes (total number ≥5) across cancer types. The point size indicates the number of affected samples. Genes are sorted by the total number of IRAVs in all cancer types. See also Supplementary Fig. 4.