Application of Raman spectroscopy and machine learning for determination of pro-toxicant activation in CYP2E1-expressing cells

Abstract

Developing a convenient, accurate, and cost-effective analytical method for the detection of Cytochrome P450 (CYP) mediated drug bioactivation remains a challenge. The present study proposes a method using Raman spectroscopy (RS) combined with machine learning algorithms to classify CYP2E1-expressing cells that were subjected to acetaminophen (APAP) metabolic activation. Raman spectra obtained from the cells were subjected to dimensionality reduction using principal component analysis (PCA), and machine learning algorithms were employed for further classification. Four models, i.e., PCA-k-nearest neighbors (KNN), PCA-support vector machine (SVM), PCA-logistic regression (LR) and PCA-random forest (RF), were established and effectively classified APAP-treated cells with or without CYP2E1 expression. The PCA-SVM model exhibited the highest accuracy, with a value of 94.49%. Feature analysis revealed that signals at 1440, 999, 645, 618, 1089, 1340, 1655, 1319, 716, 1123, 847, 745, 823, and 956 cm⁻¹ were the key features that were responsible for the classification, which indicated alterations in lipids, nucleic acids, and amino acids in the cells. This study established the feasibility of combining RS and machine learning for the detection of CYP2E1 activity, offering a promising platform for rapid drug metabolism studies.