Fig. 5.

Overview of the algorithm. Reads are taken in random order and iteratively aligned to the reference. After each alignment, the reference sequence is updated according to the learning rate w, which is proportional to the normalised edit distance between the read and the reference. In this case, there is one substitution between the reference of the read; the read has a G with Phred quality score of 15 while the reference is T. One deletion and one insertion are treated thanks to a persistence vector. The persistence value p_∙ indicates the tendency of a base to be inserted or deleted at each position in the reference. This value can trigger indels update in the reference when it goes beyond a threshold