Answers are grounded in this paper and cite their sources.
Accurate drug–target interaction prediction plays an important role in drug discovery and repositioning. However, existing methods often struggle to fully exploit the complementary information between molecular structures and sequence representations. To address this issue, we propose MFECL-DTI, a dual-branch prediction framework integrating graph-based structural encoding, multiscale sequence feature extraction, cross-branch contrastive learning, and adaptive gated fusion. Specifically, the graph branch employs residual graph convolutional networks to extract structural features from drug molecular graphs and protein contact graphs, while the sequence branch combines multiscale convolutional neural networks with a Transformer to capture local patterns and contextual dependencies. Furthermore, a bidirectional contrastive learning mechanism is introduced to align the representations of the same drug–target pair across different branches. An adaptive gating mechanism then dynamically integrates complementary features for interaction prediction. Experiments on BioSNAP, DrugBank, and C.elegans achieve AUROC values of 0.925, 0.915, and 0.994, respectively, with corresponding AUPRC values of 0.927, 0.913, and 0.994. Comparative and ablation experiments demonstrate the effectiveness of the proposed framework in improving DTI prediction performance.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: