Deep learning-driven multiplexed mycotoxin detection via color-size encoded microbead imaging

Area of Science:

• Analytical Chemistry
• Biosensing
• Food Science

Background:

• Mycotoxins are significant food safety hazards, with co-contamination posing complex risks.
• Existing detection methods often lack high-throughput capabilities for multiple mycotoxins.

Purpose of the Study:

• To develop a deep learning-driven platform for homogeneous, high-throughput analysis of four mycotoxins.
• To create an innovative 2D-encoding strategy for simultaneous mycotoxin detection.

Main Methods:

• Utilized an aptamer-recognition-triggered DNAzyme walker-hybridization chain reaction (Dz-HCR) cascade for signal amplification on microbeads.
• Implemented a color-size 2D-encoding strategy using two sizes of microbeads and two fluorophores.
• Employed the YOLOv11 deep learning model for rapid image decoding and analysis.

Main Results:

• Achieved high sensitivity (<1 pg/mL) and wide linear ranges for mycotoxin detection.
• Demonstrated excellent specificity and desirable spiked recovery rates (85.6%-114.0%) in food matrices.
• Confirmed practical applicability through consistency with HPLC-MS/MS analysis.

Conclusions:

• The developed platform offers a promising, efficient strategy for high-throughput mycotoxin detection.
• This approach significantly improves data processing efficiency for food safety monitoring.
• The method addresses the challenge of synergistic toxicity from co-contaminated mycotoxins.