XLPFE: A Simple and Effective Machine Learning Scoring Function for Protein-Ligand Scoring and Ranking

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Accurate prediction of protein-ligand binding affinities is crucial for structure-based computer-aided drug design.
• Machine learning (ML) based scoring functions (SFs) show promise but often struggle with transferability to new datasets.
• Existing ML-SFs performance is highly dependent on the similarity between training and test sets.

Purpose of the Study:

• To develop a novel scoring function (SF) with improved performance and transferability for predicting protein-ligand binding affinities.
• To integrate diverse features for a more robust and generalizable ML-based SF.
• To evaluate the new SF against established methods on various protein-ligand complex datasets.

Main Methods:

• Developed XLPFE, a new scoring function using extreme trees (ET) machine learning model.
• Combined features including energy terms (X-score, AutoDock Vina), ligand properties, and protein sequence-related information.
• Validated XLPFE performance on datasets beyond the Comparative Assessment of Scoring Functions (CASF) benchmark.

Main Results:

• XLPFE demonstrated superior scoring and ranking power compared to X-score, AutoDock Vina, ΔvinaXGB, PSH-ML, and CNN-score.
• The developed SF exhibited enhanced transferability across diverse protein-ligand complex structures.
• XLPFE showed particular effectiveness with metalloenzymes and offered faster computation speeds.

Conclusions:

• XLPFE offers improved accuracy, speed, and transferability for predicting protein-ligand binding affinities.
• The novel feature integration and ML approach enhance generalization capabilities.
• XLPFE is a promising tool for broad applications in structure-based drug design.