| 引用本文： |
-
林艳梅,曹爱清,彭昱忠.一种基于去噪自编码器融合相似度的药物-靶标相互作用预测方法[J].广西科学,2024,31(5):842-853. [点击复制]
- LIN Yanmei,CAO Aiqing,PENG Yuzhong.A Drug-Target Interaction Prediction Method Based on Denoising Auto-encoders and Similarity[J].Guangxi Sciences,2024,31(5):842-853. [点击复制]
|
|
| 摘要: |
| 基于机器学习预测潜在药物-靶标相互作用(Drug-Target Interaction,DTI) 的方法是一个具有竞争力的研究主题，但当前相关的预测方法和模型在特征学习方面尚有较大的发展空间。本研究基于无监督学习思想提出了一个结合去噪自编码器和分子相似度非线性计算方式的药物-靶标相互作用预测方法。该方法通过去噪自编码器学习和构建药物-靶标相互作用对的特征，并在此基础上融入药物-药物、靶标-靶标之间的相似信息以增强药物-靶标特征的丰富度，从而提高模型的预测能力。在Enzymes、Ion channels、GPCRs和Nuclear receptors等4个基准数据集的比较实验结果表明，本研究所提出的模型显著优于PPAEDTI、AutoDTI++、CMF、Bi-PSSM、ESBoost、CNNDTI、NFSPDTI和EFMSDTI等8个较先进模型，并与另一先进模型aSDAE相当。可见，本研究所提出的模型提高了药物(化合物)与靶标相互作用的预测性能，可为新药研发和药物重新定位提供更优的药物-靶标相互作用预测支持。 |
| 关键词: 药物-靶标相互作用 深度学习 去噪自编码器 新药研发 药物重定位 |
| DOI：10.13656/j.cnki.gxkx.20240919.004 |
| 投稿时间：2022-09-18修订日期：2023-03-16 |
| 基金项目:国家自然科学基金项目(62262044)和广西自然科学基金项目(2023GXNSFAA026027)资助。 |
|
| A Drug-Target Interaction Prediction Method Based on Denoising Auto-encoders and Similarity |
|
LIN Yanmei1, CAO Aiqing1, PENG Yuzhong1,2
|
| (1.Guangxi Key Laboratory of Human-machine Interaction and Intelligent Decision, Nanning Normal University, Nanning, Guangxi, 530001, China;2.Artificial Intelligence Research Institute, Guangxi Academy of Sciences, Nanning, Guangxi, 530007, China) |
| Abstract: |
| The method of predicting potential drug-target interactions based on machine learning is a competitive research topic,but the current related prediction methods and models still have great room for development in feature learning.Based on the idea of unsupervised learning,a drug-target interaction prediction method combining denoising autoencoder and nonlinear calculation of molecular similarity is proposed in this study.This method learns and constructs the features of drug-target interaction pairs by denoising autoencoder.On this basis,the similarity information between drug-drug and target-target is integrated to enhance the richness of drug-target features,so as to improve the prediction ability of the model.The comparative experimental results on four benchmark datasets including Enzymes,Ion channels,GPCRs and Nuclear receptors show that the proposed model is significantly better than the eight more advanced models including PPAEDTI,AutoDTI + +,CMF,Bi-PSSM,ESBoost,CNNDTI,NFSPDTI and EFMSDTI,and is comparable to another advanced model aSDAE.It can be seen that the model proposed in this study improves the prediction performance of Drug-Target Interaction (DTI),and can provide better drug-target interaction prediction support for new drug development and drug repositioning. |
| Key words: drug-target interactions deep learning denoising autoencoder new drug development drug repositioning |
