Multiclass Drug Classification Using ML on High-Performance Liquid Chromatography (HPLC) Profiles

Abstract

High-Performance Liquid Chromatography (HPLC) plays a key role in pharmaceutical and
metabolomic analysis. It separates compounds in detail based on their physical and chemical
properties. Yet, making sense of complex chromatographic results to group compounds
remains challenging. This research suggests a machine learning approach to classify drug
compounds into multiple groups. It uses engineered features taken from HPLC
chromatograms. The team processed a selected dataset of over 1,600 chromatographic runs.
These runs showed a wide range of pharmaceutical compound types. From this data, they
extracted features based on retention. These included peak count highest absorbance, entropy,
and area under the curve. They sorted compounds into nine main groups like Amino Acids,
Drugs, Bioactives, and Inorganics. They tested several classifiers such as Random Forest,
Support Vector Machine, and deep neural networks. The Random Forest model preformed
best. It reached over 99% accuracy in training and 72% accuracy in testing across all groups.
This beat traditional models. The suggested method demonstratus that to combine HPLC
profiles with ML techniques. This allows for automatic scalable, and meaningful
classification. This work helps improve drug profiling, quality control, and compound
tracking in pharmaceutical and biomedical fields