BCL::Mol2D-a robust atom environment descriptor for QSAR modeling and lead optimization

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Fragment-based molecular fingerprints are crucial in computer-assisted drug discovery.
• Atom environment (AE) descriptors, like Molprint2D, are widely used for compound screening.
• Molprint2D showed strong performance in identifying active compounds across various targets.

Purpose of the Study:

• Introduce BCL::Mol2D descriptors as an advancement over Molprint2D.
• Evaluate the performance of BCL::Mol2D in drug discovery tasks.
• Demonstrate the utility of BCL::Mol2D's reversibility for model interpretability and lead optimization.

Main Methods:

• Developed BCL::Mol2D descriptors based on a universal AE library.
• Compared BCL::Mol2D with Molprint2D on nine diverse PubChem datasets.
• Trained artificial neural networks with dropout using both descriptor sets.
• Combined BCL::Mol2D with Reduced Short Range descriptors.
• Visualized 'pharmacophore maps' using BCL::Mol2D's reversible property.

Main Results:

• BCL::Mol2D descriptors outperformed Molprint2D on nine PubChem datasets.
• Artificial neural networks trained on BCL::Mol2D showed up to 26% improvement in logAUC.
• BCL::Mol2D demonstrated reversibility, allowing decomposition of predictions to substructures.
• A modest improvement was observed when BCL::Mol2D was combined with Reduced Short Range descriptors.
• Demonstrated visualization of pharmacophore maps for kinase inhibitors.

Conclusions:

• BCL::Mol2D represents a significant improvement over Molprint2D for computer-assisted drug discovery.
• The reversibility of BCL::Mol2D enhances machine learning model interpretability and guides lead optimization.
• BCL::Mol2D is a valuable tool for identifying and optimizing drug candidates, particularly for targets like serine/threonine kinases.