文献类型：期刊文献

中文题名：基于多模态特征融合的药物靶标亲和力预测

英文题名：Drug-Target Affinity Prediction Based on Multi-Modal Feature Fusion

作者：仝凌瑞[1];李建华[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2026

卷号：52

期号：1

起止页码：118

中文期刊名：华东理工大学学报(自然科学版)

外文期刊名：Journal of East China University of Science and Technology

收录：;北大核心:【北大核心2023】；

语种：中文

中文关键词：药物靶标亲和力预测;药物研发;多模态;特征融合;图神经网络

外文关键词：drug-target affinity prediction;drug development;multimodal;feature fusion;graph neural network

摘要：药物-靶标结合亲和力是评估药物与靶标相互作用强度的关键指标。目前大多数药物靶标亲和力预测方法仅关注药物或靶标的单一模态特征,未能充分挖掘多模态信息的互补性及其在提升预测性能方面的潜在价值。针对这个问题,提出了一种基于多模态特征融合的药物靶标亲和力预测模型(Multi-Model Feature Fusion for Drug-Target Affinity Prediction,MMFDTA)。该模型对药物分子采用了包括分子指纹、分子图以及ChemBERTa预训练嵌入在内的多模态信息,对靶标蛋白采用了蛋白质序列、氨基酸残基接触图以及ProtBERT预训练编码嵌入。在此基础上,模型设计了层级特征融合架构,实现药物与靶标多模态特征之间的深度交互融合。实验结果表明,该模型在Davis和KIBA数据集上优于其他基线方法,验证了所提出的多模态融合策略的有效性。
Drug-target binding affinity is a key metric for evaluating the strength of interaction between drugs and their targets.Currently,most drug-target affinity prediction methods focus on single-modal features of either the drug or the target,failing to fully exploit the complementary nature of multi-modal information and its potential value in enhancing prediction performance.To address this issue,we propose a drug-target affinity prediction model based on multi-modal feature fusion(MMF-DTA).The model incorporates multi-modal information for drug molecules,including molecular fingerprints,molecular graphs,and ChemBERTa pre-trained embeddings,and for target proteins,it uses protein sequences,amino acid residue contact maps,and ProtBERT pre-trained embeddings.Based on these features,the model adopts a hierarchical feature fusion architecture to enable deep interaction and fusion between the drug and target multi-modal features.Experimental results demonstrate that our model outperforms other baseline methods on the Davis and KIBA datasets,validating the effectiveness of the proposed multi-modal fusion strategy.