Climbs: Assessing Carbohydrate-Protein Interactions through a Graph Neural Network Classifier Using Synthetic Negative Data

Carbohydrate-protein interactions are essential for biological processes, such as cellular signaling and metabolism, and represent a large pool of untapped targets for diagnostics and therapeutics. However, current design and prediction methods fail to accurately evaluate the affinity and specificity of proteins for carbohydrates such as glucose and galactose. Here, we describe a machine learning classifier, named CLIMBS, as a novel evaluation method for protein-carbohydrate interactions and train it on crystal structures and synthetic data from unsuccessfully designed structures to effectively assess whether carbohydrate-protein complexes represent realistic, native-like structures. Compared to other methods, CLIMBS has outstanding accuracy and excellent carbohydrate specificity, supported by high AUROC and MCC values, subsecond runtime per sample, minimal bias toward either negative or positive samples, and can be employed to improve the selection of successful docking and design models of carbohydrate-protein complexes.