Machine Learning-Assisted Multiplexed Fluorescence-Labeled miRNAs Imaging Decoding for Combined Mycotoxins Toxicity Assessment

Abstract

Mycotoxins, particularly deoxynivalenol (DON) and zearalenone (ZEN), are common food contaminants that frequently co-occur in grains, posing significant health risks. This study proposed a multiplexed detection platform for simultaneous quantification and imaging of three microRNAs (miRNAs) integrated with machine learning to evaluate the combined toxicity of DON and ZEN. Based on Exonuclease III-assisted signal amplification, highly sensitive fluorescent molecular beacon probes (MBs) targeting miR-21, miR-221, and miR-27a were developed, achieving remarkable detection limits of 0.18 pM, 0.22 pM, and 0.21 pM, respectively. The MBs were efficiently delivered into cells via liposome-mediated endocytosis, enabling simultaneous intracellular imaging of the three miRNAs. By integrating machine learning algorithms, including linear discriminant analysis and principal component analysis, with RGB values extracted from cellular fluorescence images, a robust analytical platform was established for classifying miRNA expression patterns induced by various DON/ZEN concentrations. A highest single agent model was subsequently constructed to evaluate the combined toxicity, revealing that ZEN exhibited antagonistic effects on DON at low doses but synergistic effects at high doses. This sensitive and multiplexed detection method demonstrates a strong correlation between miRNA expression profiles and DON/ZEN toxicity, providing an innovative analytical tool for multicomponent toxicity assessment.