WACA-DTA: Water-Aware Geometric Biases for Structure-Conditioned Drug-Target Affinity Prediction

khan huang, Chang Li

Keywords: drug-target affinity prediction, structure-conditioned learning, water-mediated interactions, hydration field, geometric bias, cold-start

Abstract:

Drug-target affinity (DTA) prediction is central to
virtual screening, hit prioritization, and lead opti-
mization because it estimates binding strength
before costly experiments. However, current
structure-aware DTA models often treat pocket
geometry and solvent information as auxiliary
features, leaving atom-residue matching weakly
constrained by 3D locality, orientation, and water-
mediated contacts. We present WACA-DTA, a
structure-conditioned affinity model that refor-
mulates interface matching as pose-conditioned
atom-residue cross-attention with factorized di-
rect, geometric, and hydration-mediated logits.
Distance/orientation and HydraProt-derived hy-
dration cues are injected as additive logit-level
priors rather than post hoc features. Under a fixed
pair-pose protocol and shared structural prepro-
cessing, WACA-DTA improves over a matched
pocket-aware baseline on Davis and KIBA across
drug, target, and pair affinity cold-start splits,
while controlled ablations across Davis, KIBA,
and PDBbind indicate that the gains are most con-
sistent when pair-specific geometry and hydration
are injected directly into the interaction logits.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 120