Deep learning-driven discovery of anti-metastatic phytochemicals targeting MMP-1 in breast cancer via advanced contrastive learning and structural attention.

Matrix metalloproteinase-1 (MMP-1) is a key enzyme that drives extracellular matrix degradation and facilitates breast cancer progression, invasion, and metastasis. This abnormal MMP-1 activity is linked to worse patient outcomes and also improved progression of tumor, which promotes critical therapeutic target. However, recent deep learning approaches are utilized still unable to fully capture molecular structures, cross-domain molecular-protein interactions, and interpretable predictive features. These challenges are addressed by incorporating a novel Contrastive Learning-based Molecular-Protein Deep Kernel Learning (CLM-DKL) in this research to solve constraints during the high-throughput virtual screening of phytochemicals targeting MMP-1. This process is effectively fine-tuned via the Stellar Oscillation Optimizer (SOO). Moreover, the Structure-Enhanced Cross-Interaction Graph Attention Network (SECI-GAT) produces embeddings in a hierarchical manner that capture both intra-molecular structure and molecular-protein interactions. To improve the prediction of molecules with more structurally informative area-focused progress and the alignment of different representations of information, a Multi-View Contrastive Learning (MVCL) with both attentions, such as Structural Entropy Guided Attention (SEGA) and Encoder Attention Fusion (EAF), are utilized. The uncertainty-aware molecular-protein affinity prediction by CLM-DKL, and atom-residue-level contribution scores provided by the proposed Deep Learning Important FeaTures (DeepLIFT) for biological interpretability. Evaluation outcomes achieve ROC-AUC of 0.91, 0.88, 0.90, PR-AUC of 0.87, 0.84, 0.86, for ChEMBL, PubChem BioAssay, and BindingDB, respectively, and F1 scores up to 0.94, demonstrating strong predictive power, stability, and generalization. This integrated framework provides a robust, interpretable, and structurally informed pipeline for discovering potent MMP-1 inhibitors.