DTBNet: Multimodal 3D Geometric Perception with Hierarchical Attention for Robust Drug-Target Interaction and Binding Affinity Prediction.

DTBNet improves binding affinity prediction on Davis, KIBA, and PDBBind, reducing mean squared error by up to 19.8% and increasing Pearson correlation by 8.0% while staying optimal in cold-start and novel pairing scenarios.

• Why it matters: Drug discovery needs accurate drug-target interaction and binding affinity predictions, but prior models often miss key 3D physicochemical interactions and generalize poorly when drugs, targets, or pairings are unseen.
• What they did: DTBNet builds heterogeneous embeddings from protein 3D geometric features and multilevel drug representations, then uses atom–residue cross-attention followed by group-level multi-head self-attention to integrate local and global interaction signals.
• The result: On GPCR targets, independent validation yields RMSE 0.783, and attention/visualization aligns with experimentally resolved binding conformations, supporting robust, interpretable structure-based virtual screening.

AI summary of the abstract · may contain errors · not medical advice · read the paper.