GlycoWeave keeps sugar chemistry intact during docking

GlycoWeave lets a sugar molecule move into a protein’s binding region while preserving its chemical identity. The 27 October preprint treats the molecule as a constrained structure, rather than freely moving every atom.[1]

Glycans are chains of sugar units that can branch. The positions of their chemical groups and the way units connect affect which proteins recognise them. As a 2022 Essentials of Glycobiology chapter explains, these molecules also move through many conformations in solution; binding often selects a small subset.[2]

The model takes a glycan sequence and protein structure. It changes overall position, orientation and selected bond-rotation angles, while retaining connectivity, stereochemistry and ring geometry. A neural network progressively adjusts those movements using information about individual atoms, sugar units and the neighbouring protein. Training combines experimental complexes with structures refined by physical models.[1]

In the 698-complex RecentSet benchmark, every top-ranked GlycoWeave prediction passed the authors’ checks for glycan identity, plausible internal geometry and reasonable separation from the protein. These are validity checks, distinct from recovering the experimentally observed binding pose.[1]

Docking ordinarily produces a ranked collection of candidate arrangements, called poses. Structural biologists can then filter candidates using experimental information, such as contacts identified by nuclear magnetic resonance or changes caused by mutations. The glycobiology chapter describes this as a way to connect computed arrangements with measured interactions. Chemically plausible candidates give that investigation a coherent molecular starting point.[2]

Report an error or suggest a correction