IO & interoperability¶

Going in and out of AnnNet. Three native formats and two backend bridges.

In [1]:

import tempfile
from pathlib import Path
from annnet import AnnNet

G = AnnNet(directed=True)
G.add_vertices(['A', 'B', 'C'])
G.add_edges('A', 'B', edge_id='e1', weight=1.0)
G.add_edges('B', 'C', edge_id='e2', weight=2.0)
G.attrs.set_vertex_attrs('A', label='alpha')
G.attrs.set_edge_attrs('e1', confidence=0.9)

tmp = Path(tempfile.mkdtemp(prefix='annnet_tuto_io_'))
print('temp dir:', tmp)

import tempfile from pathlib import Path from annnet import AnnNet G = AnnNet(directed=True) G.add_vertices(['A', 'B', 'C']) G.add_edges('A', 'B', edge_id='e1', weight=1.0) G.add_edges('B', 'C', edge_id='e2', weight=2.0) G.attrs.set_vertex_attrs('A', label='alpha') G.attrs.set_edge_attrs('e1', confidence=0.9) tmp = Path(tempfile.mkdtemp(prefix='annnet_tuto_io_')) print('temp dir:', tmp)

temp dir: /tmp/annnet_tuto_io_3r0vbcwm

Native .annnet format¶

Lossless, manifest-aware, fastest for AnnNet round-trip.

In [2]:

from annnet import read, write

write(G, tmp / 'g.annnet')
G_native = read(tmp / 'g.annnet')
print('round-trip nv/ne:', G_native.shape)

from annnet import read, write write(G, tmp / 'g.annnet') G_native = read(tmp / 'g.annnet') print('round-trip nv/ne:', G_native.shape)

round-trip nv/ne: (3, 2)

Parquet¶

Column-stored, language-agnostic. Good when you want to inspect the graph tables outside Python.

In [3]:

from annnet import to_parquet, from_parquet

to_parquet(G, tmp / 'g_parquet')
G_parquet = from_parquet(tmp / 'g_parquet')
print('round-trip nv/ne:', G_parquet.shape)

from annnet import to_parquet, from_parquet to_parquet(G, tmp / 'g_parquet') G_parquet = from_parquet(tmp / 'g_parquet') print('round-trip nv/ne:', G_parquet.shape)

round-trip nv/ne: (3, 2)

JSON¶

Human-readable. Useful for tiny graphs and debugging.

In [4]:

from annnet import to_json, from_json

to_json(G, tmp / 'g.json')
G_json = from_json(tmp / 'g.json')
print('round-trip nv/ne:', G_json.shape)

from annnet import to_json, from_json to_json(G, tmp / 'g.json') G_json = from_json(tmp / 'g.json') print('round-trip nv/ne:', G_json.shape)

round-trip nv/ne: (3, 2)

NetworkX bridge¶

G.nx.backend() returns the cached underlying networkx.MultiDiGraph (or MultiGraph for undirected). Use it with any networkx algorithm.

In [5]:

import networkx as nx

nxG = G.nx.backend()
print('nx type:', type(nxG).__name__)
print('nodes:', sorted(nxG.nodes()))
print('shortest_path A→C:', nx.shortest_path(nxG, 'A', 'C'))

import networkx as nx nxG = G.nx.backend() print('nx type:', type(nxG).__name__) print('nodes:', sorted(nxG.nodes())) print('shortest_path A→C:', nx.shortest_path(nxG, 'A', 'C'))

nx type: MultiDiGraph
nodes: ['A', 'B', 'C']
shortest_path A→C: ['A', 'B', 'C']

igraph bridge¶

G.ig.backend() returns the cached underlying igraph.Graph. to_igraph(G) additionally returns a manifest dict; round-trip via from_igraph(igG, manifest) is lossless.

In [6]:

from annnet import to_igraph, from_igraph

igG, manifest = to_igraph(G)
print('ig vcount/ecount:', igG.vcount(), igG.ecount())

G_via_ig = from_igraph(igG, manifest)
print('round-trip nv/ne:', G_via_ig.shape)

from annnet import to_igraph, from_igraph igG, manifest = to_igraph(G) print('ig vcount/ecount:', igG.vcount(), igG.ecount()) G_via_ig = from_igraph(igG, manifest) print('round-trip nv/ne:', G_via_ig.shape)

ig vcount/ecount: 3 2
round-trip nv/ne: (3, 2)

scverse bridge (AnnData)¶

For single-cell workflows there's to_anndata / from_anndata. The edge attribute table becomes obs, the vertex attribute table becomes var, the incidence matrix becomes a sparse X.

from annnet import to_anndata, from_anndata

adata = to_anndata(G)
G_back = from_anndata(adata)

See the use-case notebooks for end-to-end examples with a real AnnData (Kang 2018 PBMC).

---

Next: 04 — Slices.