QSAR modelling of integrin antagonists using enhanced Bayesian regularised genetic neural networks

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Bioinformatics

Background:

• Integrin antagonists are crucial for treating various diseases.
• Accurate modeling of their inhibition activity is essential for drug development.
• Existing methods may struggle with complex, nonlinear relationships in molecular data.

Purpose of the Study:

• To develop an improved computational model for predicting integrin antagonist activity.
• To identify key molecular descriptors that govern the inhibitory effects of biphenylalanine derivatives.
• To enhance the optimization capabilities of Bayesian regularised genetic neural networks.

Main Methods:

• A hybrid genetic algorithm (HGA) combining pattern search (PS), simulated annealing, and threshold acceptance with BRGNN.
• Development of artificial neural network (ANN) models with 8-3-1 architecture.
• Monte Carlo cross-validation and scrambling techniques for model validation and sensitivity analysis.

Main Results:

• The HGA effectively improved BRGNN's ability to escape local minima and identify important variables.
• Validated models achieved Q² values of 0.75 for α₄β₇ and 0.74 for α₄β₁ inhibitory activities.
• Identified key molecular descriptors: 3D-Morse for α₄β₇, and Randic shape index, Burden matrix eigenvalue, and rotatable bonds for α₄β₁.

Conclusions:

• The proposed HGA technique is a powerful tool for optimizing complex systems in drug discovery.
• The developed ANN models provide reliable predictions of integrin antagonist activity.
• Specific molecular descriptors significantly influence the modulatory activities of integrin antagonists.