FreezeNet: A Lightweight Model for Enhancing Freeze Tolerance Assessment and Genetic Analysis in Wheat.

Area of Science:

• Agricultural Science
• Plant Breeding
• Computational Biology

Background:

• Freeze injury severely impacts wheat yield, necessitating freeze-tolerant varieties.
• Traditional freeze tolerance assessment is subjective and lacks accuracy.
• Accurate phenotyping is crucial for identifying genetic factors controlling freeze tolerance.

Purpose of the Study:

• To develop an objective and accurate image-based phenotyping method for quantifying wheat freeze injury.
• To identify genetic loci associated with wheat freeze tolerance using genome-wide association studies (GWAS).
• To demonstrate the utility of identified genetic loci in enhancing wheat freeze resistance.

Main Methods:

• Developed FreezeNet, a lightweight deep learning model for image-based freeze injury quantification.
• Extracted freeze tolerance traits (VA, GVA, YVF, mHue) from smartphone images of 220 wheat accessions.
• Conducted GWAS using extracted traits to identify quantitative trait loci (QTLs) for freeze tolerance.

Main Results:

• FreezeNet accurately quantified freeze injury, with traits strongly correlated to visual scores and exhibiting high heritability.
• Identified eleven significant QTLs for freeze tolerance, including eight novel loci.
• Enhanced freeze resistance in wheat germplasm by integrating four identified QTLs.

Conclusions:

• FreezeNet provides an advanced tool for objective assessment of wheat freeze injury.
• Identified genetic loci can be effectively utilized in breeding programs to develop freeze-tolerant wheat varieties.
• This research facilitates the development of climate-resilient wheat for stable agricultural production.