SynerDTI: a synergistic deep learning framework for drug-target interaction prediction via global feature coordinated attention mechanism.

To accelerate drug discovery and repurposing, many studies increasingly rely on computational methods to predict Drug-Target Interactions (DTIs). However, existing models often face challenges related to deep feature fusion and accurate interaction modeling between drugs and targets, which limit their predictive performance. To solve these problems, this paper proposes a novel deep neural network framework referred to as SynerDTI. It incorporates an Inception Convolutional Neural Network, which employs multi-scale convolutional techniques to efficiently extract local and global features, allowing to increase the modeling capabilities of multi-scale protein structures. In addition, a Residual Graph Convolutional Network is used to accurately capture complex topological features within drug molecular graphs by propagating and aggregating information from neighboring nodes. Moreover, a Global Feature Collaborative Attention mechanism is designed for feature fusion. It combines residual connections with attention operations in a parallel architecture. This synergistic mechanism allows the model to focus on key structural regions and learn richer interactive representations of drug and target features. Furthermore, the Kolmogorov-Arnold Network is introduced in the prediction phase to increase the nonlinear modeling capacity of the framework, which allows to increase its prediction accuracy and generalization ability. Experiments are finally conducted to compare SynerDTI with fifteen state-of-the-art models. The obtained results show that it consistently outperforms them on many benchmark datasets. These results demonstrate the high prediction efficiency of SynerDTI, highlighting its potential as a significant advancement in the computational drug discovery domain.