The receptor-like neural network for modeling corticosteroid and testosterone binding globulins

Area of Science:

• Computational chemistry
• Molecular modeling
• Artificial intelligence in drug discovery

Background:

• Steroid binding globulins like CBG and TBG are crucial for hormone transport and bioavailability.
• Understanding ligand interactions with these proteins is vital for drug design and development.
• Existing methods for predicting binding affinities can be computationally intensive or lack predictive accuracy.

Purpose of the Study:

• To develop a novel neural network-based method for simulating corticosteroid and testosterone binding globulin (CBG, TBG)-ligand interactions.
• To create a computational model capable of predicting ligand binding affinity with high accuracy.
• To investigate the separation of electrostatic and shape effects in TBG affinity modeling.

Main Methods:

• Utilized molecular modeling to obtain geometric and charge data for 31 steroid molecules.
• Employed a self-organizing map (SOM) trained on high-affinity compound coordinates to create a receptor site topology template.
• Developed an unsupervised neural network for pattern recognition and similarity assessment between reference and tested ligands.

Main Results:

• Achieved a good correlation between the neural network's output signals and experimental CBG affinities.
• Developed a modified procedure for TBG affinity modeling that effectively separates electrostatic and shape contributions.
• Demonstrated high predictive power by maintaining analogy to real receptor site processes.

Conclusions:

• The presented neural-net method offers a powerful and accurate approach for simulating CBG and TBG ligand interactions.
• The model's ability to mimic receptor site topology and separate binding effects enhances its predictive capabilities.
• This computational strategy holds significant promise for accelerating drug discovery and development for steroid-related targets.