Predicting Dog Phenotypes from Genotypes.

Area of Science:

• Genomics and Canine Science
• Bioinformatics and Machine Learning Applications

Background:

• Understanding the genetic basis of canine phenotypes is crucial for selective breeding and genetic research.
• Predicting observable characteristics (phenotypes) from DNA variations (genotypes) is a key challenge in animal genetics.

Purpose of the Study:

• To predict canine breed, height, and weight using genotype data.
• To evaluate various machine learning techniques for genotype-phenotype prediction in dogs.
• To determine the minimum number of single nucleotide polymorphisms (SNPs) required for accurate prediction.

Main Methods:

• Analysis of dog genotypes (DNA variations) to predict phenotypes (observable traits).
• Application of linear and non-linear classification and regression models, including neural networks.
• Systematic evaluation of prediction accuracy using varying subsets of genomic data, from 20 SNPs to ~200k SNPs.

Main Results:

• Linear methods proved effective for breed classification, matching or exceeding non-linear approaches.
• Non-linear methods were superior for predicting dog height and weight.
• Accurate prediction of canine phenotypes was achieved using as few as 0.5% of randomly selected SNPs (992 SNPs).
• Dog breed classification achieved 50% balanced accuracy with a minimal 0.02% of SNPs (40 SNPs).

Conclusions:

• Genotype data, even with a limited number of SNPs, can reliably predict key canine phenotypes.
• The choice of statistical method (linear vs. non-linear) depends on the specific phenotype being predicted.
• This study highlights the efficiency of using targeted genomic information for predicting dog characteristics.