Quantitative structure-activity relationship studies of progesterone receptor binding steroids

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Cheminformatics

Background:

• Quantitative Structure-Activity Relationships (QSARs) are crucial for drug discovery.
• Effective QSAR models rely on the selection of relevant molecular descriptors.
• Various feature selection and mapping methods exist, each with potential strengths and weaknesses.

Purpose of the Study:

• To compare the performance of different feature selection and mapping methods for QSAR model development.
• To evaluate the predictive accuracy of various 2D QSAR protocols.
• To identify the most effective method for constructing QSAR models for steroid binding affinity.

Main Methods:

• The study employed and compared four distinct feature selection and mapping routines: forward stepping regression (FSR), genetic function approximation (GFA), generalized simulated annealing (GSA), and genetic neural network (GNN).
• QSAR models were constructed using a dataset of steroids with known in vitro binding affinity to the progesterone receptor.
• The predictive performance of each model was assessed using both training and test compound sets.

Main Results:

• The genetic neural network (GNN) protocol demonstrated superior predictive performance compared to FSR, GFA, and GSA for the tested 2D QSAR models.
• Analysis of descriptors selected by GNN aligned with established structure-activity relationships (SARs) for the steroid series.
• GNN provided the most accurate predictions for steroid binding affinity among the evaluated methods.

Conclusions:

• The genetic neural network (GNN) method is a highly effective approach for developing predictive 2D QSAR models.
• The selection of appropriate descriptors, facilitated by methods like GNN, is critical for successful QSAR formulation.
• The findings support the utility of GNN in drug discovery and lead optimization for steroid-based compounds.