Using gene expression data to predict the quality of pork

By Marie Damon, INRA

 

The research in module V of Q-PorkChains aims at identifying markers of pig meat quality, to evaluate quality of pork and control its variability.

 

Our experimental device was designed in order to induce a high variability in pork quality, using two pure breeds: Large White (conventional, n=20) and Basque (local breed giving rise to high sensory quality pork and pork products, n=30), reared in different rearing systems.

 

Considering that a biomarker of meat quality is not only correlated with the trait studied but should also predict it, we performed statistical analyses to select an informative prediction model for each of 8 quality traits:

• meat colour
• lightness: L*
• redness: a*
• ultimate pH
• drip loss
• intramuscular fat content
• shear force
• sensory tenderness 
• juiciness

Two linear statistical methods (regression and sparse PLS) and a nonparametric method (random forest), were applied on expression data obtained on 50 Longissimus muscle samples using a custom Agilent muscle microarray (15K).

Results: Within each statistical model, the number of factors (or predictors) chosen was the ones that minimized the predicted residual sum of squares (PRESS). Afterward, for each trait, the model with minimum PRESS was selected as the best predictive model.

 

Thus, lists of 4 up to 11 predictors that explain between 36% and 86% of variability were found for 7 out of the 8 meat quality traits considered (see table below), whereas selection of predictors of meat juiciness was unsuccessful.

 

Perspectives: Confirmation of predictors by real time RT-PCR on the same samples is in progress. Thereafter, predictors will have to be validated on samples from other pork chains before being considered and developed into actual tools to predict pork quality traits.

 

Meat Quality Trait	Number of genes for prediction	R2 (%)	Model used	Correlation between prediction and measurement (%)
Tenderness	7	86	Linear Regression	91
pHu	6	60	Random Forest	79
a*	11	52	Random Forest	75
L*	9	52	Random Forest	75
Shear force	8	51	Random Forest	74
Imf	6	44	Random Forest	67
Drip	4	36	Random Forest	60

_{Models for prediction of pork quality}