Deep Learning Algorithms Enabled Visual Detection of Anthrax Biomarkers by Mn₃O₄ Nanozyme-Based Colorimetric Sensor...

This study develops an innovative approach that integrates a colorimetric sensor array (CSA) composed of phenylalanine-modified Mn₃O₄ nanozymes with advanced algorithms, aiming to detect the anthrax biomarker 2,6-pyridine dicarboxylic acid (2,6-PDA) and six other structural analogs. The nanozymes exhibit tunable oxidase-like activity, catalyzing the oxidation of the chromogenic substrate 3,3',5,5'-tetramethylbenzidine (TMB) to produce blue-colored oxTMB, resulting in quantifiable color changes modulated by the PDAs' interactions. With the assistance of multivariate statistical analysis, the CSA effectively discriminates among the seven structural analogs, enabling the rapid quantitative detection of 2,6-PDA with a detection limit (LOD) of 0.015 ± 0.002 μM and allowing for visual monitoring of anthrax biomarker levels in fetal bovine serum. Moreover, by employing the deep learning YOLOv8 algorithm to visually analyze and train the linear discriminant analysis (LDA) plots obtained from the CSA, the sensor array is able to automatically classify and detect 2,6-PDA and its six structural analogs without the need for visual identification. The results show that the model's mean average precision (mAP) on the validation data set reaches 0.98-0.99, and its average confidence on the training data set can reach 0.90-0.93 (on a scale of 0 to 1). Compared to traditional manual analysis, YOLOv8 algorithm assistance significantly reduces detection time and labor costs while maintaining the accuracy of detection results, providing technical support for the application of sensor arrays in complex environments.