Improving drug-target interaction prediction by integrating a fragment method with transformer and bilinear attention network

Deep learning-based methods have made significant strides in identifying drug-target interactions (DTIs), which is a critical component in the drug discovery process. However, these methods tend to rely on global features among drugs and proteins, often fail to capture the long-distance dependencies that are essential for modeling. Thereby overlooking the complex interaction between drug fragments and amino acids. In this paper, we propose a fragment method based on transformer and bilinear attention network to predict DTIs. Specifically, our approach begins by branch chain mining and category fragment mining methods to fragment drugs and proteins, thereby obtaining their different substructures. Then utilize a transformer to learn the features of drug fragments and a convolutional neural network to learn the features of protein fragments, respectively. Combine with a bilinear attention network to explicitly capture the local, pairwise interactions between drug fragments and amino acids. Experiments on two benchmark datasets demonstrate that our method achieves significantly improved performance by comparing it with six state-of-the-art baselines.