False-Negative Recovery

Background

The original NetworKIN used network context (STRING) only as a false-positive filter applied on top of motif predictions. Any kinase–substrate pair the motif step missed was permanently excluded.

pynetworkin introduces a false-negative recovery step to rescue such pairs.

Algorithm

A pair (kinase, substrate) is recovered if:

1. It was not scored by the motif step (absent from motif_scores).
2. Its context score ≥ CONTEXT_RECOVERY_THRESHOLD (default: 0.6).

The context score is computed as:

context_score = 1 / (1 + shortest_path_distance)

where shortest_path_distance is from the Floyd–Warshall all-pairs matrix on the STRING network.

A threshold of 0.6 corresponds to an effective path length ≤ 0.67.

Output Columns

Recovered predictions include:

Column	Value
recovered	True
recovery_method	"context_proximity"
motif_score	-1.0 (sentinel, not scored)

Tuning

Adjust the threshold via:

from pynetworkin.recovery import CONTEXT_RECOVERY_THRESHOLD
# or set programmatically before calling recover_false_negatives()

Higher threshold → fewer but more confident recoveries. Lower threshold → more recoveries, higher false-positive rate.