Fig. 1: Pipeline of SCISSOR for detecting shape changes at a single gene.

a Main steps of SCISSOR are illustrated with four typical shape change examples using a toy gene: an intact example (gray); exon skipping (pink); intron retention (yellow); deletions (blue). The toy gene consists of three exons highlighted by colored background and two introns by a white background. In each example, short reads are represented by colored bars and the bars joined by the solid lines indicate the bridging reads spanning two separate genomic regions. Abnormally spliced reads from each scenario result in aberrant shapes in base-level RNA-seq expression profile as indicated by brown dashed boxes. In each coverage figure, the x axis represents genomic coordinates of the gene and the y axis represents the read-depth, i.e., the number of reads aligned to each nucleotide. b The normalized coverage accentuates the aberrant features by eliminating the common structure shared by the majority of samples. c The scores obtained by projecting the normalized coverage matrix onto each of the most outlying directions corresponding to each example are shown with the kernel density estimates. The colored point in each scatter plot indicates the outlyingness score corresponding to each example. d Main steps in the pipeline are outlined. RNA-seq coverage for a single gene is extracted from a BAM file for each subject, and a data matrix is constructed by collecting coverage data from all subjects. The data matrix is then pre-processed through transformation, exclusion of degraded samples, and normalization. The proposed statistical procedure is applied to the normalized data, providing the most outlying directions, outlyingness statistics, and identified shape outliers.