Fig. 5

Hybrid error correction outperforms long read self-correction. With 50x coverage of short-read data, hybrid error correction produces a more complete assembly than self-correction, even at twice the minimum recommended long-read coverage