ProtCB-bind: Protein-carbohydrate binding site prediction using an ensemble of classifiers

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Proteins and carbohydrates are fundamental biomolecules essential for numerous biological processes, including immune response and energy storage.
• Identifying protein-carbohydrate binding regions is crucial for understanding these vital interactions.
• Current computational methods require improvement for accurate prediction of these interactions.

Purpose of the Study:

• To develop and validate ProtCB-Bind, a novel computational model for predicting protein-carbohydrate interactions.
• To leverage advanced machine learning and Natural Language Processing (NLP) techniques for enhanced prediction accuracy.
• To provide an accessible tool for researchers studying protein-carbohydrate binding.

Main Methods:

• ProtCB-Bind employs an ensemble of machine learning classifiers with a common averaging approach for predictions.
• The model is trained using sequence-based, evolutionary-based protein features, and amino acid physicochemical properties.
• Transformer-based NLP features are integrated to improve predictive performance.

Main Results:

• ProtCB-Bind was systematically optimized for classifier and feature combinations.
• The model demonstrated superior performance compared to existing state-of-the-art predictors like SPRINT-CBH, StackCB-Pred, and StackCB-Embed.
• An approximate 3% improvement in overall performance was achieved on a benchmark dataset.

Conclusions:

• ProtCB-Bind represents a significant advancement in computational prediction of protein-carbohydrate interactions.
• The model's enhanced accuracy and integration of NLP features offer valuable insights into biomolecular recognition.
• The source code is publicly available, facilitating further research and application in the field.