A Reliable Method for Identification of Antibiotics by Terahertz Spectroscopy and SVM

Given the extensive use of antibiotics at present, the identification of antibiotics and production quality monitoring are of high importance. However, conventional antibiotic identification methods have a low sensitivity and a long detection time. Here, we propose an identification method that combines terahertz (THz) spectroscopy and chemometric technology. THz time-domain spectroscopy (THz-TDS) was performed for sixteen types of antibiotics, including β-lactam, cephalosporins, macrolides, and tetracyclines. The absorption spectra within the frequency range of 0.2–1.5 THz were calculated. For dimensionality reduction, principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE) were implemented, respectively. The data after dimensionality reduction were input into a support vector machine (SVM). The model parameters were optimized through grid search (GS), genetic algorithm (GA), and particle swarm optimization (PSO) methods, and the optimal identification results were obtained after comparison across these methods. Experiments indicate a differentiation of the THz absorption spectra among the sixteen types of antibiotics. After dimensionality reduction, the training time of the model significantly decreased. The use of the t-SNE-PSO-SVM model achieved the highest average accuracy on the prediction set, which was 99.91%. Thus, our study does not only confirm that the t-SNE-PSO-SVM model proves to be a reliable method for antibiotics identification, but also confirms that the combination of THz-TDS and chemometric pattern recognition has great potential for drug detection.