OmniPath table ingestion¶

OmniPath is a database of molecular interactions. Its Python client, omnipath-client, builds an AnnNet graph from an interaction table with omnipath_client.to_annnet. AnnNet itself does not connect to a knowledge base: the client of the knowledge base returns AnnNet objects. This notebook uses a small local table in the OmniPath format, so the result is always the same.

To run this notebook, create the environment from omnipath-table-ingestion.yml. See Reproduce a tutorial.

Build a graph from a local table¶

The table is a Polars data frame. to_annnet takes the column names for the endpoints, the edge IDs, the direction and the weight. The other listed columns become edge attributes.

In [1]:

import omnipath_client as oc
import polars as pl

interactions = pl.DataFrame(
    {
        'source': ['EGF', 'EGFR', 'EGFR', 'EGFR', 'RAS', 'MEK'],
        'target': ['EGFR', 'RAS', 'RAS', 'GRB2', 'MEK', 'ERK'],
        'interaction_id': [
            'EGF_EGFR',
            'EGFR_RAS_primary',
            'EGFR_RAS_secondary',
            'EGFR_GRB2_complex',
            'RAS_MEK',
            'MEK_ERK',
        ],
        'is_directed': [True, True, True, False, True, True],
        'curation_score': [0.95, 0.88, 0.63, 0.76, 0.82, 0.79],
        'consensus_direction': [1, 1, 1, 0, 1, 1],
        'source_database': [
            'omnipath',
            'omnipath',
            'literature',
            'complexportal',
            'pathwayextra',
            'kinaseextra',
        ],
    }
)

anet = oc.to_annnet(
    interactions,
    source_col='source',
    target_col='target',
    edge_id_col='interaction_id',
    directed_col='is_directed',
    weight_col='curation_score',
    edge_attr_cols=['consensus_direction', 'source_database'],
)

print('shape:', anet.shape)
anet.attrs.table('edges', derived=True).select(
    ['edge_id', 'source', 'target', 'effective_weight', 'source_database']
)

import omnipath_client as oc import polars as pl interactions = pl.DataFrame( { 'source': ['EGF', 'EGFR', 'EGFR', 'EGFR', 'RAS', 'MEK'], 'target': ['EGFR', 'RAS', 'RAS', 'GRB2', 'MEK', 'ERK'], 'interaction_id': [ 'EGF_EGFR', 'EGFR_RAS_primary', 'EGFR_RAS_secondary', 'EGFR_GRB2_complex', 'RAS_MEK', 'MEK_ERK', ], 'is_directed': [True, True, True, False, True, True], 'curation_score': [0.95, 0.88, 0.63, 0.76, 0.82, 0.79], 'consensus_direction': [1, 1, 1, 0, 1, 1], 'source_database': [ 'omnipath', 'omnipath', 'literature', 'complexportal', 'pathwayextra', 'kinaseextra', ], } ) anet = oc.to_annnet( interactions, source_col='source', target_col='target', edge_id_col='interaction_id', directed_col='is_directed', weight_col='curation_score', edge_attr_cols=['consensus_direction', 'source_database'], ) print('shape:', anet.shape) anet.attrs.table('edges', derived=True).select( ['edge_id', 'source', 'target', 'effective_weight', 'source_database'] )

[2026-10-12 00:34:33] [DEBUG] [pkg_infra.session        ] ▸ Using app logger: default

[2026-10-12 00:34:33] [DEBUG] [pkg_infra.session        ] ▸ Session time initialized: utc=2026-10-11 23:34:33.568800+00:00, local=2026-10-12 00:34:33.568801+01:00, timezone=BST

[2026-10-12 00:34:33] [DEBUG] [pkg_infra.session        ] ▸ Location lookup skipped

[2026-10-12 00:34:33] [INFO ] [pkg_infra.session        ] ▸ Creating new session...

[2026-10-12 00:34:33] [INFO ] [pkg_infra.session        ] ▸ Session has been created

shape: (6, 6)

Out[1]:

shape: (6, 5)

edge_id	source	target	effective_weight	source_database
str	str	str	f64	str
"EGF_EGFR"	"EGF"	"EGFR"	0.95	"omnipath"
"EGFR_RAS_primary"	"EGFR"	"RAS"	0.88	"omnipath"
"EGFR_RAS_secondary"	"EGFR"	"RAS"	0.63	"literature"
"EGFR_GRB2_complex"	"EGFR"	"GRB2"	0.76	"complexportal"
"RAS_MEK"	"RAS"	"MEK"	0.82	"pathwayextra"
"MEK_ERK"	"MEK"	"ERK"	0.79	"kinaseextra"

Add analysis context as slices¶

A slice is a named selection of edges. The high_confidence slice keeps the interactions with a weight of 0.85 or more.

In [2]:

rows = list(anet.attrs.table('edges', derived=True).iter_rows(named=True))
edge_label = {row['edge_id']: f'{row["source"]} -> {row["target"]}' for row in rows}
high_confidence = [row['edge_id'] for row in rows if row['effective_weight'] >= 0.85]
anet.slices.add('high_confidence')
anet.slices.add_edges('high_confidence', high_confidence)

print(
    'high-confidence interactions:',
    [edge_label[eid] for eid in sorted(anet.slices.edges('high_confidence'))],
)

rows = list(anet.attrs.table('edges', derived=True).iter_rows(named=True)) edge_label = {row['edge_id']: f'{row["source"]} -> {row["target"]}' for row in rows} high_confidence = [row['edge_id'] for row in rows if row['effective_weight'] >= 0.85] anet.slices.add('high_confidence') anet.slices.add_edges('high_confidence', high_confidence) print( 'high-confidence interactions:', [edge_label[eid] for eid in sorted(anet.slices.edges('high_confidence'))], )

high-confidence interactions: ['EGFR -> RAS', 'EGF -> EGFR']

Draw the network¶

The edge labels show the source database of each interaction.

In [3]:

from annnet.utils import plotting

plotting.plot(
    anet,
    backend='graphviz',
    show_edge_labels=True,
    edge_label_keys=['source_database'],
)

from annnet.utils import plotting plotting.plot( anet, backend='graphviz', show_edge_labels=True, edge_label_keys=['source_database'], )

Out[3]:

The graph from the client keeps the confidence, the source database and every analysis context in one object. The client is responsible for the origin of the table. AnnNet is responsible for what the analysis does with the graph.