Deep learning-enhanced development of innovative antioxidant liposomal drug delivery systems from natural herbs

Free radical-mediated oxidative damage to biological macromolecules, such as DNA and proteins, significantly contributes to cellular ageing. Antioxidants play a crucial role in mitigating this process by neutralizing reactive oxygen species (ROS) and reducing DNA damage. Traditional herbal medicines are of strong interest as potential sources of antioxidants due to their rich diversity of bioactive components. In this study, we developed a two-stage BERT-based framework trained on 587 experimentally confirmed antioxidants and 983 inactive compounds. The optimized model effectively screened a broad range of potential antioxidant compounds from a library of 2882 natural herbal compounds, achieving an accuracy improvement of approximately 20% over traditional machine learning models. Molecular docking simulations and <i>in vitro</i> experiments consistently validated the antioxidant capacity of the selected compounds. Additionally, incorporating three representative compounds into a liposomal delivery system not only enhanced <i>in vivo</i> bioavailability, but also mitigated oxidative stress injury after kidney acute ischemia/reperfusion. This was achieved by up-regulating antioxidant-related genes in target organs as well as ROS scavenging. Our findings highlight the potential of integrating deep learning-based compound screening with an engineered liposomal delivery platform in the research of oxidative stress and aging.