Improved method of structure-based virtual screening based on ensemble learning.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Virtual screening is crucial for drug design but faces challenges in accurately predicting binding affinity and improving hit rates.
• Current docking software scoring functions have limitations in accurately assessing protein-ligand interactions.
• Enhancing the efficiency and accuracy of structure-based virtual screening remains a key objective in computational drug discovery.

Purpose of the Study:

• To develop a novel target-specific virtual screening method, ENS-VS, to enhance hit identification in drug design.
• To improve the accuracy of predicting compound activity by integrating diverse predictive features.
• To provide a more effective computational tool for structure-based drug discovery pipelines.

Main Methods:

• Developed ENS-VS, an ensemble learning approach combining protein-ligand interaction energy terms and ligand structure vectors as descriptors.
• Integrated Support Vector Machine, Decision Tree, and Fisher Linear Discriminant classifiers within the ENS-VS framework.
• Validated ENS-VS performance on established benchmark datasets (DUD-E and DEKOIS) against existing virtual screening tools.

Main Results:

• ENS-VS demonstrated a 6-fold higher enrichment factor (EF) at 1% compared to Autodock Vina.
• On DUD-E datasets, ENS-VS achieved statistically significant higher mean EF 1% (52.77) and AUC (0.982) than SIEVE-Score (42.64, 0.912).
• On DEKOIS datasets, ENS-VS also outperformed SIEVE-Score with higher mean EF 1% (29.73) and AUC (0.793) versus (25.56, 0.765).

Conclusions:

• ENS-VS offers a significant advancement in structure-based virtual screening, outperforming current state-of-the-art methods.
• The combination of interaction energy terms and structural features, coupled with ensemble learning, enhances predictive accuracy.
• The open-source availability of ENS-VS facilitates its adoption and further development in the drug discovery community.