PCIM-DTA: pairwise conditional interaction modeling for drug-target affinity prediction under cold-start scenarios.

PCIM-DTA achieves state-of-the-art performance in drug-target affinity prediction under various cold-start scenarios, including cold-drug and cold-target conditions.

• Why it matters: Accurate prediction in cold-start situations is crucial for drug discovery but remains difficult due to static interaction models that cannot adapt to new drug-target pairs.
• What they did: The method constructs pair-level interaction representations and derives a pair-specific condition vector from global features, utilizing attention, recalibration, and graph message passing across datasets like Davis and BindingDB-Kd.
• The result: This approach outperforms existing models, enabling more reliable affinity predictions in challenging cold-start contexts and supporting further drug discovery efforts.