Deep learning-based method for predicting and classifying the binding affinity of protein-protein complexes.

Area of Science:

• Computational Biology
• Biophysics
• Bioinformatics

Background:

• Protein-protein interactions (PPIs) are fundamental to cellular functions.
• Accurate estimation of PPI binding affinity is crucial for understanding molecular recognition.

Purpose of the Study:

• To develop a deep learning model for predicting protein-protein complex binding affinity (ΔG).
• To provide an efficient computational tool for studying PPIs.

Main Methods:

• Compiled a dataset of 903 protein-protein complexes with experimental binding affinities.
• Extracted 8-20 features from sequence and predicted 3D structures for each functional class.
• Utilized feature selection and a deep learning approach for affinity prediction.

Main Results:

• Achieved a mean absolute error of 1.05 kcal/mol and a correlation of 0.79 between predicted and experimental ΔG.
• Demonstrated 87% accuracy and 0.86 F1 score in discriminating high and low affinity PPIs using 10-fold cross-validation.
• Developed a web server for predicting PPI binding affinity.

Conclusions:

• The deep learning approach offers an efficient method for predicting PPI binding affinity.
• The model provides insights into molecular recognition mechanisms in protein-protein interactions.
• The developed web server is a valuable resource for researchers studying PPIs.