Fig. 1

Illustration of CAS. The upper part shows a read and a reference. Suppose that \(t = 4\), i.e., we want to find all alignments of the read in the reference with fewer than 4 edits. There is only one such locally optimal alignment (marked as red). The middle part shows the seed extraction result with the pigeonhole principle, which splits the read into \(t = 4\) seeds. This gives many seed locations and thus many alignments. With CAS (in the lower part), we can split the read into 2 long seeds while still guarantee to find all valid mappings. The two long seeds together have a total seed frequency of 2, drastically reducing the number of alignments