Shared and distinct exonic parts in alternative paths of splicing bubbles

Abstract: Alternative splicing creates complex bubbles in splicing graphs where more than two transcript paths compete, challenging methods designed for simple binary events. We present a unified framework that compares paths using distinct exonic parts observed directly from reads. We build a GrASE splicing graph (DAG) per gene, enumerate bubbles, and quantify shared and distinct exonic parts across three comparison structures. (i) all-pairwise contrasts (ii) a multinomial n-way comparison and (iii) valid bipartitions of paths. For (iii) we introduce lower-set bipartitioning, which respects subset relations among paths by enumerating downward-closed sets in a containment graph, yielding valid two-group splits with nonempty distinguishing parts. Our test statistic is the fraction of reads mapped to distinct parts relative to distinct + shared parts, enabling differential usage across samples. Applied to genome annotations, the approach examines more bubbles than prior tools while remaining tractable and interpretable.