Concise Cascade Methods for Transgenic Rice Seed Discrimination using Spectral Phenotyping.

Area of Science:

• Agricultural Science
• Spectroscopy
• Bioinformatics

Background:

• Global regulations mandate strict traceability for genetically modified (GM) organisms in agro-food markets.
• Current detection methods for GM products are complex, time-consuming, and require specialized expertise.
• Ensuring the safety, integrity, and accurate labeling of food products necessitates reliable GM detection techniques.

Purpose of the Study:

• To develop a rapid and concise pipeline for identifying transgenic rice seeds using spectral imaging and deep learning.
• To investigate the metabolomic variations induced by specific GM traits (cry1Ab/cry1Ac gene) in rice.
• To establish an effective and efficient method for GM rice seed detection, enhancing food safety and risk assessment.

Main Methods:

• A pipeline method combining spectral imaging (near-infrared and terahertz) with a cascade deep learning model was developed.
• Metabolome composition was analyzed across three rice seed lines with the cry1Ab/cry1Ac gene to identify GM-induced variations.
• A modified guided backpropagation algorithm was employed to select characteristic wavelengths, reducing spectral data redundancy.

Main Results:

• Spectral data from different genotypes revealed regularities in GM metabolic variation.
• The cascade deep learning model achieved high accuracy: 97.04% for variety classification and 99.71% for GM status identification using terahertz spectra.
• The proposed method demonstrated viability, simplicity, and effectiveness in detecting GM rice seeds.

Conclusions:

• The developed spectral imaging and deep learning pipeline offers a rapid and accurate tool for identifying transgenic rice seeds.
• Terahertz absorption spectra proved particularly valuable for distinguishing GM traits and varieties.
• This approach holds significant potential for expedited transgenic risk assessment and improving the traceability of GM products.