| 柴瑞峰,夏瑜豪,崔新月,王苏慧,张贵军.基于图神经网络的抗体蛋白质结构优化方法[J].高技术通讯(中文),2025,35(12):1325~1336 |
| 基于图神经网络的抗体蛋白质结构优化方法 |
| An optimization method for antibody protein structure based on graph neural network |
| |
| DOI：10. 3772 / j. issn. 1002-0470. 2025. 12. 006 |
| 中文关键词: 抗体； 抗体结构优化； 图神经网络； 预训练语言模型 |
| 英文关键词: antibody, optimization of immune body structure, graph neural network, pre-training language model |
| 基金项目: |
| 作者 | 单位 | | 柴瑞峰 | （浙江工业大学信息工程学院杭州 310023 ） | | 夏瑜豪 | | | 崔新月 | | | 王苏慧 | | | 张贵军 | |
|
| 摘要点击次数: 37 |
| 全文下载次数: 29 |
| 中文摘要: |
| 抗体在免疫应答、疾病防御等方面都发挥着至关重要的作用。目前，抗体蛋白质结构预测方法对抗体的互补决定区预测仍然是一个挑战。本文设计了一种基于图神经网络的抗体蛋白质结构优化方法——图神经网络的抗体优化模型(optimization model of immune body structure based on graph neural network，GraphIR)，给定抗体的初始结构，通过抗体预训练语言模型获得初始结构的互补决定区序列表征和其他序列特征以及结构特征。然后，设计了一个针对抗体结构优化的等变图神经网络优化抗体互补决定区结构。实验结果显示，在46个抗体基准测试集上，GraphIR预测的互补决定区(complementarity determining regions，CDR)H3区域的平均均方根偏差(root mean square deviation，RMSD)为1.37?，预测精度比对标的方法ABodyBuilder、RepertoireBuilder、RosettaAntibody和DeepAb分别提升了7.45%、6.11%、7.65%和1.27%。 |
| 英文摘要: |
| Antibodies play a crucial role in immune response, disease defense, and other aspects. At present, predicting the complementary determining regions of antibodies using antibody protein structure prediction methods remains a challenge. This article proposes an optimization model of immune body structure based on graph neural network (GraphIR). Given the initial structure of the antibody, the complementary decision region sequence representation and other sequence features as well as structural features of the initial structure are obtained through antibody pre-training language models. Then, an equivariant graph neural network is designed to optimize the complementary determining region structure of antibodies. The experimental results show that on 46 antibody benchmark test sets, the average root mean square deviation (RMSD) of the CDR H3 region predicted by GraphIR is 1.37?, and the prediction accuracy is improved by 7.45%, 6.11%, 7.65%, and 1.27% compared to the benchmark methods ABodyBuilder, RepertoireBuilder, RosettaAntibody, and DeepAb, respectively. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
