cbgb

Coloured de Bruijn Graph Builder (& other methods for updating graph-edges)

This is a reference implementation for experimenting with graph algorithms on small genome de Bruin graph assemblies.

from cbgb.omfug import lnec
from cbgb import CdB, kmerise

seq = "pointy bird \ o pointy-pointy \ anoint my eyes \ anointy-nointy"

def join_walk(walk):
    return ''.join(map(lambda c: c[-1:], walk))

for k in range(4, 10):
    g = CdB(kmerise(seq, k=k)

    start = seq[:k-1]
    walk = join_walk(g.walk(start=start))
    print("{}-mer:\t{}".format(k, start[:-1] + walk))

Which yields

4-mer:	pointy-nointy \ anoint my eyes \ anointy-pointy bird \ o pointy
5-mer:	pointy-nointy \ anoint my eyes \ anointy-pointy bird \ o pointy
6-mer:	pointy-nointy \ anoint my eyes \ anointy-pointy bird \ o pointy
7-mer:	pointy \ anoint my eyes \ anointy-pointy bird \ o pointy-nointy
8-mer:	pointy bird \ o pointy-pointy \ anoint my eyes \ anointy-nointy
9-mer:	pointy bird \ o pointy-pointy \ anoint my eyes \ anointy-nointy

If we're interested in enumerating the number of possible Eulerian cycles for k,

for k in range(2, 11):
    g = CdB(kmerise(seq, k=k)):
    g.circularize()
    print(f"{k}-mer:\te^{lnec(g.adj())} cycles")

Giving:

2-mer:	e^62.73851624705093 cycles
3-mer:	e^33.7717247974044 cycles
4-mer:	e^24.99524900805808 cycles
5-mer:	e^17.317385507379875 cycles
6-mer:	e^10.514990744055561 cycles
7-mer:	e^4.564348191467836 cycles
8-mer:	e^1.3862943611198904 cycles
9-mer:	e^0.6931471805599453 cycles
10-mer:	e^0.0 cycles

Using the compress() and to_gfa() methods, we can visualise the output as a GFA file:

Monoidal edges

Integers for a multigraphs, sets for labels, dictionaries for labelled multigraphs

Eulerian walk

Linear time assembly for well behaved, unlabelled de Bruijn graphs

What's subdawg?

Extract a directed acyclic word subgraph from between two nodes

Other methods

• Enumerate Eulerian cycles with BEST theorem: docs/enumerate_cycles.ipynb
• Estimate optimal k-mer sizes with entropy/perplexity