CSM-carbohydrate

CSM-carbohydrate predicts protein-carbohydrate binding affinities and ranks docking poses to model protein–carbohydrate interactions.

Key Features:

• Machine learning algorithms: Uses machine learning algorithms to predict binding affinities for protein–carbohydrate complexes.
• Scoring function: Implements an innovative scoring function to rank docking poses.
• Curated dataset: Trained and validated on a curated dataset of 370 protein-carbohydrate complexes with experimental structural and biophysical data.
• Graph-based structural signatures: Represents protein and carbohydrate complementarity using graph-based structural signatures that capture shape and chemical properties.
• Performance metrics: Achieved Pearson's correlation coefficients of 0.72 in cross-validation (RMSE 1.58 kcal/mol) and 0.67 on an independent test set (RMSE 1.72 kcal/mol).
• Carbohydrate scope: Generalizes across mono-, di-, and oligosaccharides and can be applied to larger complexes.
• Mutation assessment: Can assess how mutations affect binding affinities.

Scientific Applications:

• Docking pose assessment: Ranking and selection of docking poses for protein–carbohydrate complexes.
• Binding affinity prediction: Predicting quantitative binding affinities for protein–carbohydrate interactions.
• Mutation impact analysis: Evaluating the effects of sequence or structural mutations on binding affinity.
• Mechanistic studies: Elucidating molecular mechanisms of protein–carbohydrate binding recognition across different carbohydrate types.

Methodology:

Trained and validated on a curated set of 370 protein-carbohydrate complexes with experimental structural and biophysical data, using graph-based structural signatures that capture shape and chemical properties, machine learning algorithms to predict binding affinities, and an innovative scoring function to rank docking poses; performance evaluated by cross-validation (Pearson 0.72, RMSE 1.58 kcal/mol) and independent testing (Pearson 0.67, RMSE 1.72 kcal/mol).