GlycanInsight: an open platform for carbohydrate-binding pocket prediction and characterization

Area of Science:

• Structural biology
• Computational chemistry
• Biochemistry

Background:

• Carbohydrate-protein interactions are crucial in biological processes but challenging to study.
• Identifying glycan-binding sites on proteins is difficult due to glycan complexity and limited computational tools.

Purpose of the Study:

• To introduce GlycanInsight, a deep learning platform for predicting carbohydrate-binding pockets on protein structures.
• To provide a tool that facilitates mechanistic studies and rational design of glycan-targeted therapeutics.

Main Methods:

• Developed a deep learning-based open platform, GlycanInsight.
• Trained and evaluated the platform on benchmark datasets of experimental and AlphaFold2-predicted protein structures.

Main Results:

• GlycanInsight achieved a Matthews correlation coefficient (MCC) of 0.63 on experimental structures, outperforming existing tools.
• The platform demonstrated robust performance on AlphaFold2-predicted structures with an MCC of 0.53.
• GlycanInsight clusters predicted residues into 3D pockets, analyzes pocket characteristics, and suggests potential ligands.

Conclusions:

• GlycanInsight offers a powerful solution for identifying carbohydrate-binding sites, overcoming previous computational limitations.
• The platform's integrated features enable detailed structural inspection, comparative analysis, and ligand prediction.
• GlycanInsight is expected to significantly advance research in glycan-protein interactions and therapeutic development.