admin管理员组

文章数量:1589797

Briefings in Bioinformatics 论文解析

现状问题：These methods are mostly based on neighbored features in sequence, and thus limitd to capture spatial information

模型：GraphPPIS

PPi位点预测转化为一个图节点分类任务（graph node classification task)

关键技术：初始残差(initial residual)和身份映射技术(identity mapping techniques)

想法：把蛋白质视为无向图，将蛋白质位点预测视为节点分类问题，整合evolutionary and structural information to construct node features and calculated pairwise amino acid distances to construct the adjacency matrix. 然后，利用初始残差和身份映射实现一个深度进化框架，从高阶氨基酸领域捕获信息。

三个数据集：

Dset_186,Dset_72,Dset_164

To ensure that the training and test set obey similar distributions in terms of interacting percentages(确保里面的PPI占比相同？）, 将他们融合

最终：335 protein chains(Train_335), 60 chain(Test_60)

Test_315 验证模型泛化性

UBtest_31 To evaluate the robustness of GraphPPIS and the impact of conformational changes on method performance. corresponding unbound structures

问题1:怎么计算pairwise amino acid distance

Node features:

two groups of amino acid features: evolutionary information(PSSM and HMM), and structural properties(DSSP)

Evolutionary information:

PSSM:position-specific scoring matrix

HMM:hidden Markov models

Structural properties:

DSSP 包括了三个结构信息

Adjacency matrix 获取方式

第一步：根据蛋白质的PDB，获得每个氨基酸的原子坐标，然后计算了所有残基队之间的欧式距离，形成了一个距离图

第二步： 将小于或等于所选cutoff 为 1，大于cutoff 为0，将该蛋白质距离图转换为邻接矩阵，custoff：14A

different strategy: to process protein distance maps into continuous matrices in which values are in the range of 0 to 1. 用公式进行标准化，如果小于或者等于cutoff

GCN with initial residual and identify mapping

 

 

本文标签： InteractionSiteproteinStructureAware

版权声明：本文标题：Structure-aware protein-protein interaction site prediction using deep graph convolutional network 内容由热心网友自发贡献，该文观点仅代表作者本人， 转载请联系作者并注明出处：https://www.elefans.com/dongtai/1728076402a1144532.html， 本站仅提供信息存储空间服务，不拥有所有权，不承担相关法律责任。如发现本站有涉嫌抄袭侵权/违法违规的内容，一经查实，本站将立刻删除。