A unified framework for drug–target interaction prediction by semantic-guided meta-path method

Drug-target interaction (DTI) prediction plays a crucial role in drug development, impacting areas such as virtual screening, drug repurposing, and the identification of potential drug side effects. Despite significant efforts dedicated to improving DTI prediction, existing methods still struggle with the challenges posed by the high sparsity of DTI datasets and the complexity of capturing heterogeneous information in biological networks. To address these challenges, we propose a unified framework for DTI prediction based on a semantics-guided meta-path walk. Specifically, we first pre-train drug and protein embeddings to capture their semantic information. This semantic information is then leveraged to guide a meta-path-based random walk on the biological heterogeneous network, generating sequences of interactions. These sequences are used to compute embedding features via a heterogeneous skip-gram model, which are subsequently fed into downstream tasks to predict DTIs. SGMDTI achieves substantial performance improvement over other state-of-the-art methods for drug–target interaction prediction. Moreover, it excels in the cold-start scenario, which is often a challenging case in DTI prediction. These results indicate the effectiveness of our approach in predicting drug-target interactions.Experimental datasets and experimental codes can be found in https://github.com/HYLPRC/SGMDTI