Issues

Polarity is often unresolved

For most hypotheses, there isn't yet enough data to say whether a compound would help or harm the disease — only the compound's effect on the shared bridge gene is known; whether modulating that bridge is itself beneficial or harmful for the disease usually isn't. A direct, curated verdict exists for about 8% of pairs (the ✓ Confirmed badge). Everywhere else, treat a strong Novelty/Strength score as a lead worth investigating, not a validated direction — read the underlying papers before drawing conclusions.

Disease-term fragmentation is mitigated, not eliminated

The same real disease can be indexed under multiple related terms (e.g. a genetic subtype and its generic parent term), which can make a well-studied connection look artificially novel under one term. Related-term evidence is now surfaced as a note when detected, but diseases aren't merged into one canonical identity, so a pair the name-matching doesn't catch can still slip through.

Compound name matching can silently miss edges

A handful of data sources match compounds by exact name. A synonym or trade name that doesn't match our stored name drops that edge entirely rather than creating a duplicate, which can undercount the Strength score for affected compounds.

Some bridge-gene interactions have no directional data

Three of the six data sources (DGIdb, DrugCentral, Open Targets) record that a relationship exists but not which direction it acts in. Polarity for those specific edges shows "no directional evidence found" even though a real interaction is known — the arrow's absence reflects a gap in that source's data, not an absence of any relationship.

The 50/50 Novelty/Strength weighting is a starting point, not a calibrated model

Combined score currently blends Novelty and Strength equally. This hasn't yet been validated against known, real-world drug-repurposing successes — it's a reasonable default, not a tuned formula.