提出了一种基于双层双向长短时记忆网络(bi-directional long short term memory，BiLSTM)和自注意力(self-attention)机制的药物-药物相互作用(drug-drug interactions，DDIs)预测方法SA-BiLSTM。首先，利用FP3指纹、MACCS指纹、Pubchem指纹和PaDEL分子描述符对药物特征信息进行提取。其次，使用套索回归(least absolute shrinkage and se-lection operator，Lasso)方法消除对分类无关的特征，并利用重复编辑最近邻(repeated edited nearest neighbors，RENN)方法对数据进行平衡处理，得到最优特征向量。最后，将最优特征向量输入结合自注意力机制和双向长短时记忆网络的分类器预测DDIs。基于五折交叉验证，同时与其它预测方法进行比较，本工作所提出的方法在两个数据集上获得较高的预测准确率。为了综合评价SA-BiLSTM的性能，对药物-药物相互作用网络进行验证。实验结果表明，SA-BiLSTM表现出优秀的预测能力，可以为DDIs的预测提供一种新的思路。
Predicting Drug-Drug Interactions with BiLSTM and Self-Attention Mechanism
In this paper,we propose a drug-drug interactions(DDIs)prediction method called SA-BiLSTM,which is based on bi-directional long short term memory(BiLSTM)and self-attention mechanism.First,the characteristic information of the drug is extracted using FP3 fingerprints,MACCS fingerprints,Pubchem fingerprints,and PaDEL molecular descriptors.Second,Lasso is applied to eliminate redundant features.Then,repeated edited nearest neighbors(RENN)method is used to balance the data to obtain the best feature vector.Fi-nally,the best feature vectors are imported into the classifier combined with self-attention and BiLSTM for predicting DDIs.SA-BiLSTM achieves high prediction accuracy on both datasets based on 5-fold cross-validation and comparison with other prediction methods.To further evaluate the predictive performance of SA-BiLSTM,the drug-drug interaction net-work is validated.The experimental results show that SA-BiLSTM achieves better predic-tion and can provide a new idea for predicting DDIs.
万方数据
维普
