Accelerating the Screening of Small Peptide Ligands by Combining Peptide-Protein Docking and Machine Learning.

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning applications

Background:

• Peptide-protein docking is crucial for drug discovery but computationally intensive.
• Accelerating small peptide ligand screening requires efficient computational methods.
• Existing methods often demand significant high-performance computing resources.

Purpose of the Study:

• To develop and validate a novel pipeline coupling machine learning (ML) and molecular docking for accelerated peptide-protein docking prediction.
• To assess the performance of various ML algorithms for classifying peptide-protein docking outcomes.
• To demonstrate the computational efficiency and accuracy of the proposed ML-driven approach for small peptide ligand screening.

Main Methods:

• Eight machine learning algorithms were evaluated for peptide-protein docking prediction.
• Light Gradient Boosting Machine (LightGBM) was selected for its computational efficiency.
• A library of 160,000 tetrapeptide ligands was screened against four viral envelope proteins using ML and molecular docking.
• The ML model was trained on 1% of the data and used to classify the remaining 99%.

Main Results:

• Light Gradient Boosting Machine (LightGBM) demonstrated superior computational efficiency compared to other ML algorithms.
• The trained LightGBM model achieved an accuracy of 0.81-0.85 and an F1-score of 0.58-0.67 in classifying peptide-protein docking performance on 99% of the data.
• The ML-coupled molecular docking pipeline achieved 90-95% concurrence with traditional methods while accelerating the process by at least 10-fold.
• The approach proved to be independent of specific molecular docking software used.

Conclusions:

• Machine learning coupled with molecular docking offers an efficient strategy for accelerating small peptide ligand screening.
• The developed pipeline effectively identifies top-performing peptides without requiring high-performance computing.
• This method provides a valuable tool for the rapid identification of potential bioactive compounds in drug discovery.