Fig. 2

Experiments to determine appropriate amounts of iterations for both DBG-based pre-correction of long reads, as performed by FMLRC, LoRDEC, and Ratatosk, and subsequent OG-based correction by HERO-OG, as per steps 1 to 6 in Fig. 1 . The x -axis indicates the different iterations. The y -axis indicates the resulting error rates. Experiments were performed on dataset containing simulated PacBio CLR reads from 3 Salmonella strains. In a and c, only FMLRC, LoRDEC, and Ratatosk were used. In b and d , 3 iterations of FMLRC, LoRDEC and Ratatosk are followed by 5 iterations of HERO-OG. Results point out that for DBG-only-based approaches, 8 iterations are optimal, while for hybrid (DBG and OG) approaches, 3 iterations of DBG, followed by 5 iterations of OG yield optimal error rates