Table 2.
Performance on 1000 Genomes Project chromosome graphs
| Model | Build | Load | Iteration rate (millions) |
||
|---|---|---|---|---|---|
| B/bp | B/bp | Node/s | Edge/s | Step/s | |
| vg | 80.2 | 77.2 | 24.6 | 2.8 | 2.9 |
| hg | 36.7 | 23.9 | 59.5 | 18.9 | 127.2 |
| og | 30.3 | 13.7 | 24.1 | 11.5 | 8.2 |
| pg | 37.6 | 3.80 | 63.7 | 4.6 | 24.3 |
| xg | 54.3 | 9.31 | 54.2 | 20.5 | 117.0 |
Note: Average build memory, load memory and iteration times for graph elements for the chromosome-level graphs built from all the variants in the 1000 Genomes Project and the GRCh37 reference genome against which the variant set was originally reported. VG requires ∼20 times as much memory to load the graphs as PackedGraph, while even the most costly libbdsg model (HashGraph) requires ∼1/3 as much memory. In these graphs, ODGI provides the lowest performance for handle iteration. However, in all other metrics, VG performs much worse than the other models. Bold indicates lowest memory usage or fastest operation.