AI药物-靶点预测-iGRLDTI论文

论文题目 iGRLDTI: an improved graph representation learning method for predicting drug-target interactions over heterogeneous biological information network

发表期刊 Bioinformatics

DTI的全称是：drug-target interactions. 药物靶点相互作用

iGRLDTI由Three steps构成:

1、HBIN construction - heterogeneous biological information network (HBIN)异质生物信息网络

A HBIN is defined as a three-element tuple, i.e. ...= {V, C, E}

其中，V包含Vd(drug) 和Vt (target)；C是一个矩阵表示drugs跟targets的生物知识；E由所有的|E|DTIs组成

Purpose： to integrate DTIs and the biological knowledge of drugs and targets, such that these different sources of information can be learned in a single context.

----  -----  ----  ---- ---  ---- ----  ---- ----- ---- ---  ---- ----  ---- ----- ---- ---  ---- ----  ---

1. V一共有549种drugs以及424个targets --- 药物以及靶点的相互作用是来自数据集DrugBankV3.0 database以及HPRD2009 databases。一共有1923个DTIs

2. Cd--是从DrugBank来的SMILES序列，通过RDKit tool处理得到cd--is considered as the feature vector of each drug

---- --- -- -- --- ---- ---- ---- ---- -- ---- ---- --- - ---  ---- ----  ---- ----- ---- ---  ---- -----

2、representation learning - 这个过程是通过an NDLS strategy来实现的. 具体来说，它允许在学习过程中自适应指定HBIN中每个生物分子的传播深度.

3、DTI prediction - iGRLDTI trains a GBDT classifier with distinguishable feature representations of drugs and targets to complete the prediction task.