基于图注意力和蛋白质语言模型的药物靶标亲和力预测  

Drug-target Affinity Prediction Based on Graph Attention Network and Protein Language Model

作　　者：任鹏 段乐乐 REN Peng;DUAN Lele(Chongqing Three Gorges Medical College,Chongqing 404100)

机构地区：[1]重庆三峡医药高等专科学校,重庆404100

出　　处：《软件》2024年第4期24-28,共5页Software

基　　金：2021年重庆三峡医药高等专科学校自然科学研究项目“基于图神经网络的药物靶点相互作用预测方法研究”(XJ2021001601)。

摘　　要：在新药研发中,药物与靶标的结合亲和力是一个关键指标,对于它的准确预测对加速药物筛选至关重要。近年来,随着深度学习技术的应用,药物与靶标的结合亲和力预测取得了显著进步,受到了广泛关注。然而,现有方法大多忽略药物的结构信息,且缺乏学习蛋白质的序列与结构之间的深层模式,导致预测性能受限。针对该问题,本文提出了一种新的方法,它结合了图注意力网络和蛋白质语言模型,以更有效地表征药物和蛋白质。本文在两个公开的数据集上进行了实验,并与其他同类方法进行了比较。实验结果显示,本文提出的方法能够显著提升预测准确率,这证明了本文提出的方法的有效性。In the development of new drugs,the binding affinity between drugs and targets is a key indicator,and accurate prediction of it is crucial for accelerating drug screening.In recent years,significant progress has been made in the prediction of drug-target binding affinity,attracting widespread attention due to the application of deep learning technologies.However,most existing methods tend to overlook the structural information of drugs and lack in-depth learning of the patterns between protein sequences and structures,resulting in limited predictive performance.To address this issue,this paper introduces a novel approach that integrates graph attention network with protein language model to represent drug and protein more effectively.Experiments conducted on two public datasets,compared with other similar methods,demonstrate that the proposed method significantly enhances prediction accuracy,thereby proving its effectiveness.

关 键 词：图注意力网络 蛋白质语言模型 药物靶标亲和力 深度学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] R9[自动化与计算机技术—控制科学与工程]