KANSARL Fusion Transcript Detection Kit

The patent segments the RNA-seq read mapping process into distinct phases: initial rapid filtering of reads using k-mer based signatures, followed by focused alignment only for reads that pass the filter. This segmentation allows the system to achieve both high speed (through rapid filtering of the majority of reads) and high accuracy (through detailed alignment of candidate reads), resolving the contradiction between speed and accuracy in fusion transcript detection

The patent applies different quality levels of processing to different subsets of reads. High-speed, lower-compute processing is applied to all reads initially, while high-accuracy, high-compute processing is applied only to the small subset of reads that pass the filtering stage. This local quality approach ensures that computational resources are concentrated where they are most needed, achieving both speed and accuracy in the overall detection process

The patent performs preliminary actions by pre-computing and storing k-mer based signatures for all reads before the actual fusion detection analysis. These pre-computed signatures enable rapid filtering in the main analysis phase without requiring complex real-time computations, thus improving detection coverage while managing computational complexity through advance preparation

The patent introduces k-mer based signatures as an intermediary representation between the raw RNA-seq reads and the final fusion transcript identification. This intermediary layer simplifies the comparison and matching processes, enabling comprehensive analysis of novel fusion transcripts while reducing the direct computational complexity of aligning and comparing full-length reads