Highly accurate carbohydrate-binding site prediction with DeepGlycanSite

Area of Science:

• Biochemistry and Structural Biology
• Computational Biology and Bioinformatics
• Drug Discovery and Development

Background:

• Carbohydrates are vital biomolecules regulating physiological and pathological processes.
• The complexity of carbohydrates hinders experimental identification of protein-binding sites.
• Understanding carbohydrate-protein interactions is crucial for therapeutic development.

Purpose of the Study:

• To develop a computational model for accurate prediction of carbohydrate-binding sites on protein structures.
• To overcome the challenges posed by carbohydrate diversity and complexity in identifying interaction sites.
• To facilitate the study of carbohydrate-protein interactions for biological and therapeutic insights.

Main Methods:

• Introduction of DeepGlycanSite, a deep learning model utilizing a deep equivariant graph neural network with transformer architecture.
• Incorporation of geometric and evolutionary features of proteins into the model.
• Validation through integration with a mutagenesis study on a G-protein coupled receptor.

Main Results:

• DeepGlycanSite accurately predicts carbohydrate-binding sites on protein structures.
• The model significantly outperforms existing state-of-the-art methods in predicting diverse carbohydrate-binding sites.
• The study identified the guanosine-5'-diphosphate-sugar-recognition site of a G-protein coupled receptor.

Conclusions:

• DeepGlycanSite is a valuable tool for predicting carbohydrate-binding sites.
• The model offers insights into the molecular mechanisms of carbohydrate regulation in therapeutically important proteins.
• This work advances the understanding of carbohydrate-protein interactions for potential therapeutic applications.