Predictions from deep learning propose substantial protein-carbohydrate interplay - PubMed
6 hours ago
- #protein-carbohydrate interactions
- #bioinformatics
- #deep learning
- Neural network PiCAP predicts noncovalent protein-carbohydrate binding with 90% balanced accuracy at the protein level.
- Supporting model CAPSIF2 identifies interacting residues, outperforming previous methods with a Dice coefficient of 0.57.
- Application across six proteomes suggests 35-40% of proteins bind carbohydrates, rising to 75% for extracellular and cell surface proteins.
- Predicted binders are linked to biological functions like growth factor receptor binding, inflammation, and cell adhesion.