SENET-AOP: A computational framework model for prioritizing antioxidant protein targets in drug discovery

Oxidative stress is closely associated with aging and a wide range of major diseases, including cancers, neurodegenerative disorders, and inflammatory conditions. Antioxidant proteins therefore constitute important functional biomolecules and potential therapeutic targets in drug discovery. Accurate and efficient identification of antioxidant proteins is thus of relevance for biomedical research and early-stage target exploration. To address the limitations of existing computational approaches, such as restricted dataset size, limited generalization ability, and simplistic feature representations, we constructed a high-quality benchmark dataset comprising 1144 antioxidant proteins and 2959 non-antioxidant proteins. On this basis, we propose SENET-AOP, an attention-based deep learning framework for antioxidant protein classification and target prioritization. The model integrates complementary semantic representations derived from two protein language models, ESM-2 and ProtT5, and adopts a dual-branch CNN-SENet architecture to capture local sequence patterns and global physicochemical properties, while adaptively recalibrating channel-wise feature importance. Experimental results demonstrate that SENET-AOP achieves an accuracy of 0.9360, a Matthews correlation coefficient of 0.8376, and an AUROC of 0.9721 under five-fold cross-validation. On the independent test set, the model attains an accuracy of 0.9367 and an AUROC of 0.9795, consistently outperforming other methods. Moreover, the proposed framework exhibits favorable interpretability. Collectively, SENET-AOP provides an effective and reliable tool for high-throughput identification and prioritization of antioxidant protein targets, supporting oxidative stress-related diseases research and medicinal chemistry-oriented drug discovery workflows. For user convenience, a freely accessible web server has been developed at: http://www.senetaop.com.cn/.