Principle

Compared to typical compound-protein interaction prediction methods, DeepETD fully utilize comprehensive biomedical data rather than relying solely on structural information of metabolites or proteins. It integrates multi-source biomedical data, including subcellular localization-metabolite associations, cell phenotype-metabolite associations, and disease-metabolite associations. DeepETD uses a deep learning algorithm with an attention mechanism to process high-dimensional and noisy data. Focusing on the most relevant parts of the data, DeepETD is able to capture the complex relationships between metabolites and proteins effectively. By focusing on the most relevant parts of the data, DeepETD effectively captures complex relationships between metabolites and proteins. The attention mechanism assigns weights to input data from different sources, which enhances metabolite-protein interaction prediction. All prediction results are stored in this web server named EMTDD. This database allows users to easily query the potential targets of endogenous metabolites predicted by DeepETD.