Geometric deep learning framework for de novo genome assembly

Table 2.

Results on ONT data

Data set Assembler Size (Mb) LG50 LG90 NG50 (Mb) NGA50 (Mb) Complete (%) Duplicated (%) QV # misasm structural # misasm local
CHM13 GNNome 2984 11 27 111.0 83.1 99.42 0.83 33.59 1758 1695
Raven 2907 16 48 72.6 70.3 99.51 0.70 34.04 1146 952
Flye 2851 17 60 70.0 69.5 99.50 0.69 33.91 134 123
Shasta 2884 17 72 59.7 56.3 99.50 0.55 37.30 57 190
A. thaliana GNNome 131 4 9 14.6 14.5 99.83 1.07 26.54 242 221
Raven 125 5 11 14.5 14.5 99.81 1.07 27.28 95 185
Flye 124 5 15 13.7 13.7 99.79 1.07 26.03 7 16
Shasta 115 5 12.1 12.1 99.81 0.94 25.84 16 256
  • The best-achieved results are in bold. The assemblies of GNNome and Raven were additionally polished with Racon (Vaser et al. 2017). Flye and Shasta have built-in polishing modules, so we did not perform additional postassembly polishing. Shasta did not assemble 90% of the A. thaliana genome, hence the LG90 is undefined. Full QUAST report for ONT data is given in Supplemental Table S2.

This Article

  1. Genome Res. 35: 839-849

Preprint Server