Structure-activity correlations for illicit amphetamines using ANN and constitutional descriptors

Area of Science:

• Computational Chemistry
• Toxicology
• Pharmacology

Background:

• Illicit substances pose significant public health and forensic challenges.
• Accurate identification of novel psychoactive substances is crucial for risk assessment.
• Predictive toxicology systems can aid in evaluating new chemical entities.

Purpose of the Study:

• To develop an expert system using artificial neural networks (ANNs) to predict the biological activity of novel amphetamine-like compounds.
• To classify amphetamines based on toxicological activity (stimulant vs. hallucinogenic) and differentiate them from non-amphetamine substances.
• To create a system for epidemiological, clinical, and forensic analysis of new molecular structures.

Main Methods:

• Two types of ANNs were designed: a spectral ANN using infrared (IR) spectral data and a structural ANN (CD-ANN) using constitutional descriptors (CDs).
• The spectral ANN utilized 100 key absorption intensities from normalized GC-FT-IR spectra of drugs of abuse and related compounds.
• The CD-ANN incorporated 45 computed CDs per sample, with optimizations to the training set and input variables.

Main Results:

• Both spectral and structural ANNs were trained to classify amphetamines by toxicological activity.
• The CD-ANN's efficiency was enhanced through training set optimization and an increased number of input variables.
• A comparative analysis of the performance of the spectral and structural networks was conducted.

Conclusions:

• Artificial neural networks can be effectively employed to predict the biological and toxicological activity of new substances.
• The developed expert system provides a valuable tool for the identification and classification of amphetamine-related compounds.
• This approach supports forensic, clinical, and epidemiological investigations involving novel psychoactive substances.