Section: New Results

Read against read comparison for Nanopore data

In the team, we developed two years ago seeds with errors, which allow to find all common approximate patterns with a limited number of errors. The idea behind these seeds, called ${01}^{\ast }0$ seeds, is to divide the sequence in blocks so that the distribution of errors is no longer random. This year, we have used these seeds in the context of long reads analysis. With this data, reads against reads comparison suffers from a high loss of sensitivity, because the single read error-rate is already high. Our application case is the detection of adapter sequences in ONT sequencing. We have shown that the use of these seeds instead of exact $k$-mers allowed a more accurate reconstruction of the sequences of the adapters. The method takes two steps: first the identification of $k$-mers potentially composing the adapter using a counting approach that takes into account errors in the read, and then the reconstruction of the complete sequence of the adapter with a greedy algorithm. Our results show that the seeds with errors allow to obtain accurate consensus sequences for more 80% of the samples, compared to 40% with the usual $k$-mer approach. This work was done within the ANR ASTER during the first year of the thesis of Quentin Bonenfant and was presented at the national workshop Seqbio 2018.