抗藥性是當前全球面臨的重大公共衛生挑戰之一。抗微生物肽（AMPs）被認為是應對日益嚴重的抗生素抗藥性威脅的有前景工具。然而，儘管AMPs具備許多優點，其在臨床應用中面臨一個關鍵挑戰，即對哺乳動物細胞的溶血毒性。為了克服這一挑戰，本研究引入名為EnHemo的整合模型，旨在辨識高活性抗菌肽的溶血毒性。EnHemo模型結合多種先進技術，包括極限梯度提升算法、殘差算法和遷移學習算法，並利用iFeature特徵和先進的蛋白質語言模型來提高解釋性和預測準確性。研究結果顯示，EnHemo模型在兩個數據集上分別達到了90.60%和96.43%的高準確率，顯著超越現有分類器在準確性和均衡分類任務上的表現。此外，EnHemo模型的多層次特徵整合和先進算法應用，顯示出其在實際應用中的潛力。總之，我們提出的EnHemo模型不僅能有效識別安全且具有高活性的抗菌肽，還為未來抗菌肽的設計和開發提供了重要的參考和指導。這一研究成果有望推動AMPs在臨床上的安全應用，為抗擊抗藥性威脅提供新的解決方案。;Antimicrobial resistance is one of the major public health challenges currently facing the world. Antimicrobial peptides are considered promising tools to address the growing threat of antibiotic resistance. However, despite their many advantages, AMPs face a critical challenge in clinical applications due to their hemolytic toxicity to mammalian cells. To overcome this challenge, we introduce an integrated model named EnHemo, designed to identify the hemolytic toxicity of high active AMPs. EnHemo combines multiple advanced technologies, including Extreme Gradient Boosting, residual algorithms, and transfer learning, utilizing iFeature features and advanced protein language models to enhance interpretability and predictive accuracy. The results show that EnHemo achieved high accuracy rates of 90.60% and 96.43% on two datasets, significantly outperforming existing classifiers in terms of accuracy and balanced classification tasks. Moreover, the multi-level feature integration and advanced algorithms of EnHemo demonstrate its potential in practical applications. In summary, the EnHemo model effectively identifies safe and highly active AMPs and offers important guidance for the design and development of future AMPs. This research outcome is expected to promote the safe clinical application of AMPs, providing a new solution to combat the threat of antimicrobial resistance.
