Principal component analysis coupled with artificial neural networks--a combined technique classifying small...

Area of Science:

• Forensic Chemistry
• Computational Chemistry
• Toxicology

Background:

• New psychoactive substances (NPS) frequently emerge on the black market by modifying controlled drug structures.
• Accurate identification of these novel compounds is crucial for public health and law enforcement.
• Existing methods may struggle to keep pace with the rapid evolution of illicit drug chemistry.

Purpose of the Study:

• To develop and validate expert systems for classifying forensic compounds.
• To differentiate amphetamines (stimulants/hallucinogens) from non-amphetamines using spectral data.
• To provide a reliable, fast, and accessible tool for molecular identification in forensic contexts.

Main Methods:

• Preprocessing of a spectral database combining GC-FTIR and GC-MS data from 103 forensic compounds.
• Application of Principal Component Analysis (PCA) to reduce data dimensionality.
• Development of eight Artificial Neural Network (ANN) models (PC-ANN systems) using varying numbers of principal components (PCs) as input.

Main Results:

• The optimal expert system, a PC-ANN model using 18 PCs, explained 77% of the variance.
• This best model achieved 100% sensitivity (true positive rate) and 92.77% selectivity (true negative rate).
• Comparative analysis highlighted the discriminating power of input variables for classification.

Conclusions:

• ANN-based expert systems, particularly PC-ANN models, are effective for classifying forensic compounds.
• The developed system demonstrates high accuracy in identifying amphetamine-related substances.
• This approach offers a promising solution for rapid identification of novel psychoactive substances in forensic investigations.