A Cross-Field Fusion Strategy for Drug-Target Interaction Prediction

TL;DR

This work tackles the challenge of predicting drug–target interactions for novel drugs and targets by integrating local and global protein information through a cross-field fusion strategy. The proposed SiamDTI uses a dual-channel, Siamese architecture with shared protein encoders and a bilinear attention-based fusion to derive a discriminative joint representation, achieving state-of-the-art performance on novel DTIs while remaining competitive for known DTIs. Key contributions include the PPI-enabled global protein channel, a BAN-based pairwise fusion module, and a complexity-aware design that reduces parameters via shared encoders. The approach promises improved robustness and scalability for large-scale virtual screening in drug discovery.

Abstract

Drug-target interaction (DTI) prediction is a critical component of the drug discovery process. In the drug development engineering field, predicting novel drug-target interactions is extremely crucial.However, although existing methods have achieved high accuracy levels in predicting known drugs and drug targets, they fail to utilize global protein information during DTI prediction. This leads to an inability to effectively predict interaction the interactions between novel drugs and their targets. As a result, the cross-field information fusion strategy is employed to acquire local and global protein information. Thus, we propose the siamese drug-target interaction SiamDTI prediction method, which utilizes a double channel network structure for cross-field supervised learning.Experimental results on three benchmark datasets demonstrate that SiamDTI achieves higher accuracy levels than other state-of-the-art (SOTA) methods on novel drugs and targets.Additionally, SiamDTI's performance with known drugs and targets is comparable to that of SOTA approachs. The code is available at https://anonymous.4open.science/r/DDDTI-434D.

Figures (4)

• Figure 1: For novel drugs and targes, traditional methods do not utilize the global information of target proteins, resulting in poor performance. SiamDTI achieves better performance by fusing global and local protein information. For known drugs and targes, both traditional methods and SiamDTI can achieve good results.
• Figure 2: Overview of the SiamDTI framework. SiamDTI consists of two channels: drug–target interaction channel and protein–protein interaction channel. The protein input for both channels are the same target protein. It accurately predicts DTI by dual channel information fusion.
• Figure 3: (a) Performance comparison of the methods for known drugs and targets on the three datasets.The histogram represents the average results of five randomized experiments, and the error bars are the size of the standard deviation. (b) Comparison of the SiamDTI and DrugBAN for sample representation distances of known drugs and targets on BindingDBgao2018interpretable. The horizontal coordinate represents the target protein, and the vertical coordinate denotes the representation distance of different labeled drug target pairs composed of that target. A larger average distance indicates better classification performance.
• Figure 4: The sample distribution for SiamDTI and Drugban within two scenarios.