A Random Forest Model for Peptide Classification Based on Virtual Docking Data.

Area of Science:

• Computational Biology
• Biochemistry
• Machine Learning

Background:

• Peptide affinity is vital for understanding peptide-protein interactions.
• Existing methods for evaluating peptide-receptor affinity may lack accuracy.
• Accurate affinity prediction is crucial for drug development and biological research.

Purpose of the Study:

• To develop a reliable method for assessing actual peptide affinity using virtual docking data.
• To create a free, accessible tool for researchers to evaluate peptide-receptor interactions.
• To leverage machine learning to improve the accuracy of peptide affinity prediction.

Main Methods:

• Combined actual peptide affinity data with virtual peptide-receptor docking scores.
• Utilized various machine learning algorithms, focusing on random forest (RF).
• Developed and validated RF models with different feature sets, including a four-feature model.

Main Results:

• The random forest algorithm demonstrated superior performance over other tested algorithms.
• The four-feature RF model achieved the highest accuracy (0.714) on an independent dataset.
• Identified and addressed overfitting issues in models with fewer features.
• Established a relationship between actual peptide affinity and virtual docking scores.

Conclusions:

• The developed four-feature RF model accurately predicts peptide affinity from virtual docking data.
• This free tool offers a more reliable assessment of peptide-receptor interactions.
• The findings contribute to advancing computational approaches in drug discovery and molecular biology.