New molecular surface-based 3D-QSAR method using Kohonen neural network and 3-way PLS.

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Pharmacology

Background:

• Comparative molecular field analysis (CoMFA) is a standard 3D-QSAR method but may not accurately reflect ligand-receptor interactions.
• Key molecular interactions occur near the van der Waals surface, suggesting grid points surrounding the entire molecule in CoMFA are not always relevant.

Purpose of the Study:

• To develop a more precise and realistic 3D-QSAR method by focusing on physico-chemical parameters on the molecular surface.
• To introduce a surface-based 3D-QSAR approach utilizing Kohonen neural network (KNN) and three-way partial least squares (3-way PLS).

Main Methods:

• Developed a surface-based 3D-QSAR method integrating KNN and 3-way PLS.
• Applied the method to 25 dopamine 2 (D2) receptor antagonists.
• Projected 3D van der Waals surface points to a 2D map using KNN, coded nodes with molecular electrostatic potential (MEP) values, and analyzed correlations with D2 antagonist activities using 3-way PLS.

Main Results:

• The 3-way PLS model demonstrated excellent statistics.
• Coefficients back-projected onto the van der Waals surface showed a reasonable 3D distribution.
• External validation using D-optimal designs indicated superior predictive performance of 3-way PLS compared to standard 2-way PLS.

Conclusions:

• The developed surface-based 3D-QSAR method provides a more realistic representation of ligand-receptor interactions.
• This novel approach enhances the predictive accuracy of 3D-QSAR models for drug discovery, particularly for D2 receptor antagonists.
• Three-way PLS offers improved predictive capabilities over traditional 2-way PLS in external validation scenarios.