COVER STORY • 封面專題 21 2024 UMAGAZINE 30 • 澳大新語 mechanisms, formulation development, and clinical trial results, accelerating the discovery of potential drugs. A team led by Lu Jiahong, deputy director of ICMS and associate professor in SKL-QRCM, has collaborated with partners such as the University of Oslo in Norway, Wenzhou Medical University, and MindRank (a drug discovery company). They have developed an advanced machine-learning algorithm for drug screening by combining the AI-assisted drug screening platform with cells, C. elegans, and mouse models of Alzheimer’s disease, and identified several small-molecule compounds from Chinese medicinal plants with potential for treating Alzheimer’s disease. According to Prof Lu, his team pre-trained a representation model incorporating multi-dimensional molecular information from data on 19 million small molecules sourced from two databases. Following this, the team screened 3,724 natural small molecules and eventually identified 18 for further validation. Their efforts led to the discovery of two promising compounds: the natural flavonoid kaempferol from sand ginger and the stilbenoid rhapontigenin from Gnetum cleistostachyum. ‘We found that these two compounds significantly improved neurodegenerative symptoms in mouse models of Alzheimer’s disease, reducing pathological markers such as the aggregation of amyloid plaques and tau protein, while also improving their learning and memory abilities,’ Prof Lu says. ‘This discovery lays a foundation for our future research into using mitochondrial activation strategies to treat Alzheimer’s disease.’ Supporting Drug Design Through Algorithm Platform Meanwhile, a team led by Ouyang Defang, associate professor in SKL-QRCM and ICMS, has developed a machine learning framework called DeepCSP to address the challenges of traditional crystal structure prediction. ‘The crystal structure of a drug directly affects its safety, efficacy, and production efficiency,’ says Prof Ouyang. ‘Different crystal structures also affect the absorption, stability, and efficacy of drugs in the human body. Given the complex and variable nature of Chinese medicine components, finding stable crystalline forms is often a challenge.’ According to Prof Ouyang, DeepCSP, which integrates generative adversarial networks with molecular graph convolutional networks, can rapidly generate potential crystal structures and predict their stability, and the whole process takes only a few minutes. 路嘉宏教授 Prof Lu Jiahong 研究員對GFP-mito-mCherry-Parkin HeLa細胞施用山奈酚和丹葉大黃素，驗證這兩款線粒體自噬激活劑的功效。 The researchers administered kaempferol and rhapontigenin to GFP-mito-mCherry-Parkin HeLa cells to validate their ability to induce mitophagy
RkJQdWJsaXNoZXIy NzM0NzM2