Fig. 6: Description of the Oncosplice pipeline.

A A diagram illustrating the greedy approach to constructing transcript isoforms given only a pool of splice sites with predicted missplicing such as missed or cryptic junctions. B Overview of the steps taken in the pipeline to obtain a concise quantitative description of the functional loss that a mutation induces through predicted missplicing, and the transcriptome annotations; general run-time per mutation consumes 7.73 seconds when running on an Intel Xeon Platinum 9294 CPU, with 40% of the time spent running SpliceAI inference and 60% spent predicting, generating, and grading variant transcriptomes. C Comparing two proteins using conservation scores per amino acid using an algorithm that captures the loss due to insertions and deletions. D A view of the collected data for each transcript’s isoforms, and the aggregation of functional loss scores at the transcript and gene level.