**中国科学院团队利用深度学习模型精准预测糖-蛋白质作用位点**
近日，中国科学院的一支研究团队取得重大突破，他们开发的深度学习模型DeepGlycanSite能够精准预测糖-蛋白质作用位点，效率达到传统方法的30倍。
糖类作为自然界中最丰富的有机物质，对生命过程至关重要。了解糖与蛋白质之间的相互作用对于解决生物学难题和开发新疗法至关重要。然而，这一研究领域面临着糖类分子多样性和复杂性的挑战。
DeepGlycanSite模型结合了蛋白质的几何和进化特征，并融入Transformer架构的深度等变图神经网络中。其性能大大超越了现有方法，能够准确预测各种糖类分子的结合位点。
结合诱变研究，该模型成功揭示了重要G蛋白偶联受体的鸟苷-5′-二磷酸糖识别位点。这一发现不仅突显了DeepGlycanSite在糖结合位点预测方面的巨大潜力，也为我们深入了解蛋白质糖类调节的分子机制提供了深入见解。
该研究成果以“Highly accurate carbohydrate-binding site prediction with DeepGlycanSite”为题，于2024年6月17日发表在《自然通讯》上，标志着人工智能与生物学交叉领域的一大进步。这一突破有望为药物研发、疾病治疗等提供更多精准有效的策略。
英语如下：
News Title: “Chinese Academy of Sciences Develops DeepGlycanSite Model: Precision Prediction of Glycoprotein Binding Sites”
Keywords: 1. Prediction of Glycoprotein Binding Sites
News Content: **Chinese Academy of Sciences Team Uses Deep Learning Model to Accurately Predict Glycan-Protein Interaction Sites**
Recently, a research team from the Chinese Academy of Sciences has made a significant breakthrough. Their developed deep learning model, DeepGlycanSite, can accurately predict glycan-protein interaction sites, with an efficiency of up to 30 times that of traditional methods.
Carbohydrates, as the most abundant organic substances in nature, are crucial to life processes. Understanding the interaction between carbohydrates and proteins is essential for solving biological problems and developing new therapeutic approaches. However, this research field faces challenges due to the diversity and complexity of carbohydrate molecules.
The DeepGlycanSite model combines protein geometry and evolutionary features, incorporating a deep equivariant graph neural network with a Transformer architecture. Its performance significantly surpasses existing methods, accurately predicting binding sites for various carbohydrate molecules.
Combined with mutagenic research, the model has successfully revealed the glycosyl-5′-diphosphate recognition sites of important G-protein-coupled receptors. This discovery not only highlights the enormous potential of DeepGlycanSite in carbohydrate binding site prediction but also provides deep insights into the molecular mechanisms of protein-glycan regulation.
The research results, titled “Highly accurate carbohydrate-binding site prediction with DeepGlycanSite,” were published in Nature Communications on June 17th, 2024, marking a significant advancement in the intersection of artificial intelligence and biology. This breakthrough is expected to provide more precise and effective strategies for drug development, disease treatment, and other applications.
【来源】https://www.jiqizhixin.com/articles/2024-07-01-4
Views: 3