Hierarchy neural networks as applied to pharmaceutical problems

Area of Science:

• Computational chemistry
• Pharmacology
• Artificial intelligence

Background:

• Artificial neural networks (ANNs) are increasingly utilized in pharmaceutical research.
• Hierarchy-type networks with backpropagation are common for optimization and prediction tasks.
• ANNs exhibit strong nonlinear classification and fitting capabilities.

Purpose of the Study:

• To review the fundamental operating characteristics of artificial neural networks (ANNs) in pharmaceutical applications.
• To analyze the merits and demerits of ANNs' nonlinear operations and propose remedies.
• To explore mathematical relationships between ANNs, ALS, and multiregression analysis.

Main Methods:

• Review of hierarchy-type networks and backpropagation learning.
• Analysis of ANN's nonlinear operational characteristics.
• Examination of analytical formulas for partial derivatives to understand input-output relationships.
• Application of reconstruction-learning and descriptor-mapping methods.

Main Results:

• ANNs possess significant nonlinear classification and fitting abilities.
• Mathematical relationships with ALS and multiregression are elucidated.
• Analytical derivatives provide insight into input-output dynamics.
• Reconstruction-learning identifies essential network components and information flow.

Conclusions:

• ANNs offer robust nonlinear capabilities for pharmaceutical optimization and prediction.
• Understanding the mathematical properties and applying advanced methods enhances ANN utility.
• Methods like derivative analysis and reconstruction-learning are key to interpreting complex relationships.