Class identity assignment for amphetamines using neural networks and GC-FTIR data.

Area of Science:

• Analytical Chemistry
• Forensic Science
• Computational Chemistry

Background:

• Accurate identification of amphetamines is crucial for epidemiological, clinical, and forensic investigations.
• Current methods for amphetamine identification can be time-consuming and require specialized expertise.
• Automation through computational methods offers potential for faster and more efficient analysis.

Purpose of the Study:

• To evaluate the feasibility of building neural network (NN) systems for automating amphetamine identification.
• To develop NN systems capable of distinguishing between amphetamines and nonamphetamines.
• To create a refined NN system for classifying amphetamines based on their toxicological activity (stimulant, hallucinogenic).

Main Methods:

• Development of two neural network systems: one for amphetamine/nonamphetamine discrimination, and a second for toxicological activity classification.
• Utilized GC-FTIR absorption intensities as input variables for the neural networks.
• Analyzed the spectroscopic interpretation of the 40 most important input variables.

Main Results:

• The first NN system achieved an 83.44% correct classification rate for distinguishing amphetamines from nonamphetamines.
• The second, more refined NN system achieved an 85.71% correct classification rate for classifying toxicological activity.
• Spectroscopic analysis indicated that the modeling power of input variables correlates with spectral interaction stability, not just absorption intensity.

Conclusions:

• Building neural network systems for automated amphetamine identification is feasible.
• The developed NN systems demonstrate high accuracy in classifying amphetamines and their toxicological effects.
• Variable selection for NN models should consider spectral stability, not solely absorption intensity, to maintain predictive power.