<![CDATA[Drug-target affinity (DTA) prediction is critical to drug discovery, yet traditional experimental methods are expensive and time-consuming. Existing computational approaches often struggle with limitations in representing the structural and sequential complexities of drugs and proteins, resulting in suboptimal prediction accuracy. This study proposes a novel framework integrating Graph Attention Networks (GAT) for drug molecular and motif graphs and Bidirectional Long Short-Term Memory (BiLSTM) for protein sequences. A two-sided multi-head attention mechanism is utilized to dynamically model drug-protein interactions, enhancing robustness and accuracy. This research contribution is the development of a robust computational model that improves the accuracy of DTA predictions, reducing dependency on traditional laboratory methods. The integration of structural and sequential features provides a more comprehensive representation of drug-protein interactions. The study utilizes the Davis and KIBA, a binding affinity datasets that is widely used. the proposed model achieving the lowest Mean Squared Error (MSE) of 0.3209 and 0.1864, the highest Concordance Index (CI) of 0.8646 and 0.8616, and the highest of 0.5046 and 0.6672, respectively, outperforming baseline models. In conclusion, this study showed the proposed approach as a reliable method for DTA prediction, offering a faster and more accurate alternative in the drug discovery research field. However, there are still limitations, such as high computational complexity and the GAT model still uses static attention. Future work will focus on addressing this issue, testing the model across broader datasets, and implementing additional drug and target representation for richer feature extraction.]]>
Copyrights © 2024
