Exploring methods to summarize gut microbiota composition for microbiability estimation and phenotypic prediction in...

Area of Science:

• Animal Science
• Microbiology
• Genetics

Background:

• Gut microbiota composition influences host traits.
• Quantifying microbial resemblance is crucial for genetic analyses.
• Various methods exist to create similarity matrices from microbial data.

Purpose of the Study:

• To investigate eight approaches for creating microbial similarity matrices.
• To compare their performance in estimating trait microbiability.
• To evaluate their accuracy in predicting growth and body composition traits in pigs.

Main Methods:

• Eight methods were used to create microbial similarity matrices: Linear Kernel, Polynomial Kernel, Gaussian Kernel, Arc-cosine Kernel, Bray-Curtis, Jaccard, Metric Multidimensional Scaling, and Detrended Correspondence Analysis.
• Microbiability and prediction accuracies for body weight, backfat, loin depth, and intramuscular fat were estimated in Duroc, Landrace, and Large White pigs.
• Four-fold cross-validation was employed for prediction accuracy assessment within each breed.

Main Results:

• Microbiability estimates varied significantly across methods, with Gaussian Kernel, Jaccard, Bray-Curtis, and Arc-cosine Kernel yielding higher values.
• Prediction accuracies for traits like loin depth were substantial, with Bray-Curtis, Metric Multidimensional Scaling, Linear Kernel, and Jaccard showing superior performance.
• Polynomial Kernel and Detrended Correspondence Analysis consistently demonstrated the poorest performance.

Conclusions:

• The choice of method for constructing microbial similarity matrices significantly impacts microbiability estimation and trait prediction in swine.
• Gaussian Kernel, Jaccard, Bray-Curtis, and Arc-cosine Kernel are recommended for microbiability estimation.
• Bray-Curtis, Metric Multidimensional Scaling, Linear Kernel, and Jaccard are suggested for phenotypic prediction in pigs.