Integrated deep network model with multi-head twofold attention for drug–target interaction prediction
Reliable Drug–Target Interaction (DTI) prediction is essential for drug development, drug repurposing, and understanding off-target effects in the rapidly advancing field of precision medicine. Traditional methods often fail to capture the complex relationships between drugs and their biological tar...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
AIP Publishing LLC
2025-06-01
|
| Series: | AIP Advances |
| Online Access: | http://dx.doi.org/10.1063/5.0278705 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849427918809202688 |
|---|---|
| author | Angelin Jeba P Tamilpavai G |
| author_facet | Angelin Jeba P Tamilpavai G |
| author_sort | Angelin Jeba P |
| collection | DOAJ |
| description | Reliable Drug–Target Interaction (DTI) prediction is essential for drug development, drug repurposing, and understanding off-target effects in the rapidly advancing field of precision medicine. Traditional methods often fail to capture the complex relationships between drugs and their biological targets due to the heterogeneity of biological data. To address these challenges, an integrated deep learning model with Multi-Head Twofold Attention (MHTA) has been proposed for DTI prediction. This model leverages convolutional and recurrent neural networks to extract both local and sequential features from drug molecular structures and target protein sequences. The MHTA mechanism enhances the model’s ability to focus on different aspects of drug and target features independently, effectively capturing intricate interaction patterns. Dense embeddings generated from input representations are refined using recurrent layers for long-range dependencies and convolutional layers for local patterns. The refined features undergo the MHTA process separately and are then aggregated and passed through fully connected layers to predict the probability of interaction. The model was evaluated on six benchmark datasets—DrugBank, BindingDB, STITCH, KEGG DRUG, Drug Target Common, and ChEMBL—and achieved outstanding performance with 99% accuracy, 98% precision, 97% recall, 98% F1-score, 98% AUC-ROC, and 97% specificity. These results demonstrate the model’s robustness and high predictive power, surpassing existing techniques and highlighting its potential to accelerate drug discovery by reliably identifying novel DTIs. |
| format | Article |
| id | doaj-art-28f8d3fe41a344e9b1bf1f35fb26fd3d |
| institution | Kabale University |
| issn | 2158-3226 |
| language | English |
| publishDate | 2025-06-01 |
| publisher | AIP Publishing LLC |
| record_format | Article |
| series | AIP Advances |
| spelling | doaj-art-28f8d3fe41a344e9b1bf1f35fb26fd3d2025-08-20T03:28:52ZengAIP Publishing LLCAIP Advances2158-32262025-06-01156065113065113-1910.1063/5.0278705Integrated deep network model with multi-head twofold attention for drug–target interaction predictionAngelin Jeba P0Tamilpavai G1Government College of Engineering, Tirunelveli 627007, IndiaGovernment College of Engineering, Tirunelveli 627007, IndiaReliable Drug–Target Interaction (DTI) prediction is essential for drug development, drug repurposing, and understanding off-target effects in the rapidly advancing field of precision medicine. Traditional methods often fail to capture the complex relationships between drugs and their biological targets due to the heterogeneity of biological data. To address these challenges, an integrated deep learning model with Multi-Head Twofold Attention (MHTA) has been proposed for DTI prediction. This model leverages convolutional and recurrent neural networks to extract both local and sequential features from drug molecular structures and target protein sequences. The MHTA mechanism enhances the model’s ability to focus on different aspects of drug and target features independently, effectively capturing intricate interaction patterns. Dense embeddings generated from input representations are refined using recurrent layers for long-range dependencies and convolutional layers for local patterns. The refined features undergo the MHTA process separately and are then aggregated and passed through fully connected layers to predict the probability of interaction. The model was evaluated on six benchmark datasets—DrugBank, BindingDB, STITCH, KEGG DRUG, Drug Target Common, and ChEMBL—and achieved outstanding performance with 99% accuracy, 98% precision, 97% recall, 98% F1-score, 98% AUC-ROC, and 97% specificity. These results demonstrate the model’s robustness and high predictive power, surpassing existing techniques and highlighting its potential to accelerate drug discovery by reliably identifying novel DTIs.http://dx.doi.org/10.1063/5.0278705 |
| spellingShingle | Angelin Jeba P Tamilpavai G Integrated deep network model with multi-head twofold attention for drug–target interaction prediction AIP Advances |
| title | Integrated deep network model with multi-head twofold attention for drug–target interaction prediction |
| title_full | Integrated deep network model with multi-head twofold attention for drug–target interaction prediction |
| title_fullStr | Integrated deep network model with multi-head twofold attention for drug–target interaction prediction |
| title_full_unstemmed | Integrated deep network model with multi-head twofold attention for drug–target interaction prediction |
| title_short | Integrated deep network model with multi-head twofold attention for drug–target interaction prediction |
| title_sort | integrated deep network model with multi head twofold attention for drug target interaction prediction |
| url | http://dx.doi.org/10.1063/5.0278705 |
| work_keys_str_mv | AT angelinjebap integrateddeepnetworkmodelwithmultiheadtwofoldattentionfordrugtargetinteractionprediction AT tamilpavaig integrateddeepnetworkmodelwithmultiheadtwofoldattentionfordrugtargetinteractionprediction |