HBADTI: Drug-target interaction prediction based on multi head attention and bidirectional cross attention

Jiaming Zhao
Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, PMLR 278:747-759, 2025.

Abstract

The study of drug-target interactions (DTIs) holds critical importance in the drug development process. The core challenge in DTI prediction lies in accurately capturing the features of both drugs and proteins, as well as thoroughly understanding their interaction mechanisms. In light of this, we developed an end-to-end DTI prediction model called HBADTI. The model employs graph convolutional networks to encode drug features. For protein feature extraction, we designed a dedicated feature extraction module (ESAM) that combines convolutional neural networks (CNNs) with multi-head self-attention mechanisms to effectively capture protein sequence characteristics. Subsequently, a bidirectional cross-attention network is utilized to integrate the features of both drugs and proteins, followed by a multilayer perceptron to classify unknown drug-target pairs.Comparative experimental results demonstrate that HBADTI outperforms multiple baseline methods. Ablation studies further confirm that both the bidirectional attention network and the ESAM module significantly contribute to the improvement of DTI prediction performance.

Cite this Paper

BibTeX
@InProceedings{pmlr-v278-zhao25c, title = {HBADTI: Drug-target interaction prediction based on multi head attention and bidirectional cross attention}, author = {Zhao, Jiaming}, booktitle = {Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing}, pages = {747--759}, year = {2025}, editor = {Zeng, Nianyin and Pachori, Ram Bilas and Wang, Dongshu}, volume = {278}, series = {Proceedings of Machine Learning Research}, month = {25--27 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v278/main/assets/zhao25c/zhao25c.pdf}, url = {https://proceedings.mlr.press/v278/zhao25c.html}, abstract = {The study of drug-target interactions (DTIs) holds critical importance in the drug development process. The core challenge in DTI prediction lies in accurately capturing the features of both drugs and proteins, as well as thoroughly understanding their interaction mechanisms. In light of this, we developed an end-to-end DTI prediction model called HBADTI. The model employs graph convolutional networks to encode drug features. For protein feature extraction, we designed a dedicated feature extraction module (ESAM) that combines convolutional neural networks (CNNs) with multi-head self-attention mechanisms to effectively capture protein sequence characteristics. Subsequently, a bidirectional cross-attention network is utilized to integrate the features of both drugs and proteins, followed by a multilayer perceptron to classify unknown drug-target pairs.Comparative experimental results demonstrate that HBADTI outperforms multiple baseline methods. Ablation studies further confirm that both the bidirectional attention network and the ESAM module significantly contribute to the improvement of DTI prediction performance.} }
Endnote
%0 Conference Paper %T HBADTI: Drug-target interaction prediction based on multi head attention and bidirectional cross attention %A Jiaming Zhao %B Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing %C Proceedings of Machine Learning Research %D 2025 %E Nianyin Zeng %E Ram Bilas Pachori %E Dongshu Wang %F pmlr-v278-zhao25c %I PMLR %P 747--759 %U https://proceedings.mlr.press/v278/zhao25c.html %V 278 %X The study of drug-target interactions (DTIs) holds critical importance in the drug development process. The core challenge in DTI prediction lies in accurately capturing the features of both drugs and proteins, as well as thoroughly understanding their interaction mechanisms. In light of this, we developed an end-to-end DTI prediction model called HBADTI. The model employs graph convolutional networks to encode drug features. For protein feature extraction, we designed a dedicated feature extraction module (ESAM) that combines convolutional neural networks (CNNs) with multi-head self-attention mechanisms to effectively capture protein sequence characteristics. Subsequently, a bidirectional cross-attention network is utilized to integrate the features of both drugs and proteins, followed by a multilayer perceptron to classify unknown drug-target pairs.Comparative experimental results demonstrate that HBADTI outperforms multiple baseline methods. Ablation studies further confirm that both the bidirectional attention network and the ESAM module significantly contribute to the improvement of DTI prediction performance.
APA
Zhao, J.. (2025). HBADTI: Drug-target interaction prediction based on multi head attention and bidirectional cross attention. Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, in Proceedings of Machine Learning Research 278:747-759 Available from https://proceedings.mlr.press/v278/zhao25c.html.

Related Material