Evaluation of Docking Machine Learning and Molecular Dynamics Methodologies for DNA-Ligand Systems

Area of Science:

• Computational chemistry
• Molecular modeling
• Drug discovery

Background:

• DNA is a key target for treating diseases like cancer.
• Few computational methods effectively analyze DNA-intercalating agent interactions.
• Developing accurate docking methodologies is crucial for drug design.

Purpose of the Study:

• To evaluate and compare docking methodologies for analyzing DNA-intercalating agent interactions.
• To assess the performance of AutoDock Vina, DOCK 6, and Consensus within Molecular Architect (MolAr).
• To develop a machine learning model for predicting DNA-ligand binding affinity.

Main Methods:

• Docking simulations using AutoDock Vina, DOCK 6, and Consensus implemented in MolAr.
• Ligand refinement using Parametric Method 7 (PM7).
• Validation through visual inspection, redocking, and Receiver Operating Characteristic (ROC) curve analysis.
• Machine learning model development for predicting melting temperature (ΔT_m).
• Molecular Dynamic Simulations (MD) using NAMD for selected ligands.

Main Results:

• The Consensus methodology achieved an Area Under the ROC Curve (AUC-ROC) of 0.99, outperforming AutoDock Vina (0.98) and DOCK 6 (0.88).
• A machine learning model demonstrated a strong predictive capability with an R² score of 0.84 for experimental ΔT_m values.
• Molecular Dynamic Simulations confirmed the equilibrium binding poses of selected DNA intercalators.

Conclusions:

• MolAr significantly improves docking results for DNA systems compared to standalone methods.
• MolAr is a robust methodology for studying DNA-ligand interactions.
• The developed approach is valuable for estimating experimental ΔT_m values of DNA intercalating agents, aiding drug discovery.