Slices and subgraphs¶

Slices keep multiple graph contexts in one AnnNet object. Operations materialize smaller graphs from those contexts.

In [1]:

import annnet as an

import annnet as an

Create condition slices¶

In [2]:

G = an.AnnNet(directed=True)
G.add_vertices(['A', 'B', 'C', 'D', 'E'])
G.slices.add('control', cohort='wildtype')
G.slices.add('treated', cohort='drug')

G.add_edges('A', 'B', edge_id='e1', slice='control', weight=1.0)
G.add_edges('B', 'C', edge_id='e2', slice='control', weight=2.0)
G.add_edges('B', 'C', edge_id='e2_treated', slice='treated', weight=5.0)
G.add_edges('C', 'D', edge_id='e3', slice='treated', weight=3.0)
G.add_edges('D', 'E', edge_id='e4', slice='treated', weight=1.5)
G.add_edges(
    'A',
    'E',
    edge_id='context_binding',
    directed=False,
    slice='treated',
    weight=0.6,
)

print(G.slices.summary())

G = an.AnnNet(directed=True) G.add_vertices(['A', 'B', 'C', 'D', 'E']) G.slices.add('control', cohort='wildtype') G.slices.add('treated', cohort='drug') G.add_edges('A', 'B', edge_id='e1', slice='control', weight=1.0) G.add_edges('B', 'C', edge_id='e2', slice='control', weight=2.0) G.add_edges('B', 'C', edge_id='e2_treated', slice='treated', weight=5.0) G.add_edges('C', 'D', edge_id='e3', slice='treated', weight=3.0) G.add_edges('D', 'E', edge_id='e4', slice='treated', weight=1.5) G.add_edges( 'A', 'E', edge_id='context_binding', directed=False, slice='treated', weight=0.6, ) print(G.slices.summary())

slices: 3
├─ default: 5 vertices, 0 edges
├─ control: 3 vertices, 2 edges
└─ treated: 5 vertices, 4 edges

Slice membership and algebra¶

In [3]:

edge_lookup = {}
for row in G.views.edges().iter_rows(named=True):
    arrow = '->' if row['directed'] else '--'
    edge_lookup[row['edge_id']] = f"{row['source']} {arrow} {row['target']}"
readable = lambda edge_ids: [edge_lookup[eid] for eid in sorted(edge_ids)]

union = G.slices.union(['control', 'treated'])
common = G.slices.intersect(['control', 'treated'])
treated_only = G.slices.difference('treated', 'control')

print('control:', readable(G.slices.edges('control')))
print('treated:', readable(G.slices.edges('treated')))
print('union:', readable(union['edges']))
print('common:', readable(common['edges']))
print('treated only:', readable(treated_only['edges']))

edge_lookup = {} for row in G.views.edges().iter_rows(named=True): arrow = '->' if row['directed'] else '--' edge_lookup[row['edge_id']] = f"{row['source']} {arrow} {row['target']}" readable = lambda edge_ids: [edge_lookup[eid] for eid in sorted(edge_ids)] union = G.slices.union(['control', 'treated']) common = G.slices.intersect(['control', 'treated']) treated_only = G.slices.difference('treated', 'control') print('control:', readable(G.slices.edges('control'))) print('treated:', readable(G.slices.edges('treated'))) print('union:', readable(union['edges'])) print('common:', readable(common['edges'])) print('treated only:', readable(treated_only['edges']))

control: ['A -> B', 'B -> C']
treated: ['A -- E', 'B -> C', 'C -> D', 'D -> E']
union: ['A -- E', 'A -> B', 'B -> C', 'B -> C', 'C -> D', 'D -> E']
common: []
treated only: ['A -- E', 'B -> C', 'C -> D', 'D -> E']

Draw the full context graph¶

The graph contains all context-specific edges. Slices decide which of those edges participate in a given context.

In [4]:

from annnet.utils import plotting

plotting.plot(G, backend='graphviz', show_edge_labels=True)

from annnet.utils import plotting plotting.plot(G, backend='graphviz', show_edge_labels=True)

Out[4]:

Per-slice edge weights¶

In [5]:

G.slices.add_edge_to_slice('treated', 'e2')
G.attrs.set_edge_slice_attrs('treated', 'e2', weight=8.0)

print('base e2 weight:', G.attrs.get_effective_edge_weight('e2'))
print('treated e2 weight:', G.attrs.get_effective_edge_weight('e2', slice='treated'))

G.slices.add_edge_to_slice('treated', 'e2') G.attrs.set_edge_slice_attrs('treated', 'e2', weight=8.0) print('base e2 weight:', G.attrs.get_effective_edge_weight('e2')) print('treated e2 weight:', G.attrs.get_effective_edge_weight('e2', slice='treated'))

base e2 weight: 2.0
treated e2 weight: 8.0

Materialize subgraphs¶

In [6]:

treated_graph = G.subgraph_from_slice('treated')
edge_graph = G.ops.edge_subgraph(['e2', 'e3'])
reversed_graph = G.ops.reverse()

print('treated shape:', treated_graph.shape)
print('edge-subgraph shape:', edge_graph.shape)
print('reversed e1:', reversed_graph.edge_definitions['e1'])

treated_graph = G.subgraph_from_slice('treated') edge_graph = G.ops.edge_subgraph(['e2', 'e3']) reversed_graph = G.ops.reverse() print('treated shape:', treated_graph.shape) print('edge-subgraph shape:', edge_graph.shape) print('reversed e1:', reversed_graph.edge_definitions['e1'])

treated shape: (5, 5)
edge-subgraph shape: (3, 2)
reversed e1: ('B', 'A', 'binary')

Draw the treated subgraph¶

In [7]:

plotting.plot(treated_graph, backend='graphviz', show_edge_labels=True)

plotting.plot(treated_graph, backend='graphviz', show_edge_labels=True)

Out[7]:

The treated subgraph materializes only the selected context: treated-only edges, the undirected context edge, and the slice-specific override on e2 are included in the derived graph.