AD-LSF: A Multimodal Framework for Drug-Target Interaction Prediction Based on Asymmetric Dynamic Gated Modulation and Adaptive Latent Signal Fusion.

Accurate prediction of drug-target interactions (DTIs) is fundamental to drug discovery and repurposing. Although deep learning-based multimodal integration has advanced DTI modeling, current approaches remain limited in capturing complex cross-modal nonlinear dependencies and in reconciling distributional heterogeneity across data modalities. To address these limitations, we propose a multimodal DTI prediction framework termed Asymmetric Dynamic Gated Modulation and Adaptive Latent Signal Fusion (AD-LSF), which is built upon asymmetric dynamic gating modulation and adaptive latent signal fusion. Specifically, we develop an asymmetric dynamic gating modulation module that leverages anisotropic convolution and dynamic frequency decomposition to capture directional structural dependencies and multiscale details within molecular sequences. In addition, we introduce an adaptive latent signal fusion mechanism that integrates latent signal coordination with bidirectional interactive alignment, enabling deep alignment and adaptive aggregation of heterogeneous features. Across four benchmark data sets, AD-LSF consistently outperforms state-of-the-art methods in AUROC, AUPRC, F1 score, recall, and accuracy.