Impact of Trait Measurement Error on Quantitative Genetic Analysis of Computer Vision-Derived Traits

Area of Science:

• Animal genetics
• Quantitative genetics
• Computer vision in agriculture

Background:

• Phenotyping using image or video data is common for quantitative genetic analysis.
• Pig body weight (BW) can be measured conventionally or via computer vision, introducing potential phenotyping errors.
• Previous research suggested genetic influence on trait measurement error, potentially biasing genetic analyses.

Purpose of the Study:

• To investigate the impact of trait measurement error on the quantitative genetic analysis of pig body weight (BW).
• To determine if genetic factors influence the error between conventional and image-derived BW measurements.

Main Methods:

• Compared scale-based (conventional) and image-based BW measurements.
• Estimated genomic heritability for both BW measurements and trait measurement error.
• Performed genome-wide association analysis (GWAS) for scale-based BW and trait measurement error.
• Evaluated different computer vision algorithms, including deep learning and adaptive thresholding.

Main Results:

• Scale-based and image-based BW measurements showed high agreement and similar genomic heritability estimates.
• Genomic heritability estimates for trait measurement error were consistently negligible.
• GWAS identified no overlapping genetic markers for scale-based BW and trait measurement error.
• Deep learning methods demonstrated superior performance over adaptive thresholding for image-based phenotyping.

Conclusions:

• Image-based BW phenotyping yields comparable quantitative genetic results to manual measurements.
• No evidence suggests that BW trait measurement error is influenced by genetic factors.
• Trait measurement error in pig BW is unlikely to introduce systematic bias into genetic analyses.