Analytical modeling of thiosemicarbazone metal complexes–based colorimetric array for high-throughput detection of multiple biogenic amines

• Published in: Mikrochimica acta (1966) Vol. 192; no. 6; p. 391
• Main Authors: Singh, Harupjit, Ranbir, Singh, Gagandeep, Kaur, Navneet, Singh, Narinder
• Format: Journal Article
• Language: English
• Published: Vienna Springer Vienna 01.06.2025; Springer Nature B.V
• Subjects: Aldehydes; Algorithms; Amines; Analytical Chemistry; Biogenic Amines - analysis; Characterization and Evaluation of Materials; Chemistry; Chemistry and Materials Science; Classification; Colorimetry; Colorimetry - methods; Combinatorial analysis; Combinatorial chemistry; Coordination Complexes - chemistry; Coordination compounds; Copper; Decision trees; Discriminant Analysis; Entropy (Information theory); High-Throughput Screening Assays - methods; Ligands; Limit of Detection; Microengineering; Nanochemistry; Nanotechnology; Neural networks; Neural Networks, Computer; Nickel; Principal Component Analysis; Principal components analysis; Sensor arrays; Thiosemicarbazones - chemistry; Biomimetic sensor array; Neural network; Multivariate analysis; Biogenic amines; Meat freshness; Principal component analysis
• ISSN: 0026-3672; 1436-5073; 1436-5073
• DOI: 10.1007/s00604-025-07256-0

A single polydentate ligand H6 was designed and synthesized using the azo-aldehyde as a colorimetric signaling unit and thiosemicarbazide derivative as a recognition unit. The concepts of combinatorial chemistry were implemented for the development of sensing elements of the array using a combination of single ligand H6  + multiple metallic cations (Cd(II), Cu(II), Ni(II), Zn (II), Na(I), and Ag(I)) and the array was utilized for the sensing of biogenic amines. Further, the effectiveness of the developed sensor array in distinguishing between analytes was assessed and compared using five distinct algorithms: principal component analysis (PCA), linear discriminant analysis (LDA), decision tree (DT), random forest (RF), and perceptron neural networks. The outcomes of principal component analysis concurred with the original hypothesis and established the discriminatory power of the array to detect multiple amines. Thereafter, classification of amines was performed using linear discriminant analysis and validated by leave-one-out cross-validation method, resulting in the remarkable accuracy of 98% when samples of varying concentrations were utilized, with detection limits in the range of 0.55–1.13 µM. Further, a combined principal component analysis and neural network (PCA + NN)–based algorithm was developed by using 6 PCA components as input of the neural network, having 3 hidden layers and 11 outputs for performing classification of biogenic amines. The PCA + NN algorithm outperformed all other methods and resulted in the maximum accuracy of 98.6% for successful classification of amines using the Adam optimizer and categorical cross-entropy as the loss function. Finally, the sensor array was successfully utilized for monitoring the quality of real chicken meat samples. Graphical Abstract