A comparison between 2D and 3D descriptors in QSAR modeling based on bio-active conformations

Area of Science:

• Computational Chemistry
• Cheminformatics
• Drug Discovery

Background:

• Quantitative Structure-Activity Relationship (QSAR) models are crucial in various research fields.
• Model performance relies heavily on molecular descriptors (1D, 2D, 3D, 4D).
• Previous comparisons of 2D vs. 3D descriptors were inconclusive, often due to unrepresentative ligand conformations.

Purpose of the Study:

• To investigate the impact of combining 2D and 3D molecular descriptors on QSAR model performance.
• To create a novel dataset of protein-ligand complexes with ligands in their bioactive conformations.
• To assess descriptor performance using machine learning algorithms and external validation.

Main Methods:

• Curated a dataset of 461 protein-ligand complexes from the Protein Data Bank (PDB), focusing on uniform activity data and bioactive conformations.
• Generated 2D, 3D, and combined 2D+3D molecular descriptors for each ligand.
• Built QSAR models using k-Nearest Neighbors, Random Forest, and Lasso Regression.
• Evaluated model performance on randomly and rationally derived external test sets.

Main Results:

• Models incorporating both 2D and 3D descriptors yielded significantly more accurate and robust results.
• The combination of descriptor types captured complementary molecular properties, enhancing predictive power.
• This approach proved superior to using either 2D or 3D descriptors alone.

Conclusions:

• Combining 2D and 3D molecular descriptors is a superior strategy for developing high-performance QSAR models.
• The use of bioactive conformations in descriptor calculation is critical for reliable QSAR modeling.
• This study provides a valuable resource and methodology for advancing drug discovery and chemical research.