Preprocessing of HPLC trace impurity patterns by wavelet packets for pharmaceutical fingerprinting using artificial...

Area of Science:

• Analytical Chemistry
• Chemometrics
• Pharmaceutical Analysis

Background:

• Pharmaceutical fingerprinting is crucial for quality control.
• High-Performance Liquid Chromatography (HPLC) is a standard technique for analyzing drug impurities.
• Developing efficient data preprocessing methods is essential for accurate classification.

Purpose of the Study:

• To evaluate wavelet packets (WPs) as a preprocessor for HPLC impurity profiles.
• To compare WP performance against a previous Window preprocessor for pharmaceutical fingerprinting.
• To assess the effectiveness of WPs with artificial neural networks (ANNs), KNN, and SIMCA classifiers.

Main Methods:

• HPLC data from 253 L-tryptophan samples from six manufacturers were analyzed.
• Wavelet packet (WP) decomposition using the Haar function was applied to impurity patterns.
• ANNs, KNN, and SIMCA classifiers were used with WP-derived input features (20-50 inputs).

Main Results:

• Optimal ANN performance reached 97% correct classification with 30 inputs, exceeding the Window preprocessor (93%).
• KNN classifier achieved 97% accuracy with 20 WP inputs, outperforming the Window preprocessor (85%).
• WP preprocessor showed comparable or superior performance to the Window preprocessor with fewer inputs for ANN and KNN.

Conclusions:

• Wavelet packets offer a superior preprocessor for HPLC-based pharmaceutical fingerprinting compared to the Window method.
• WP preprocessors enable higher classification accuracy with reduced input features for ANNs and KNN.
• This approach enhances the efficiency and effectiveness of classifying pharmaceutical products based on impurity profiles.