By Gary Hartley

Machine learning can aid animal breeding by sifting genomic data — but flaws remain

Machine learning algorithms have “great potential” for use in predicting what genomic markers will make the biggest impact in animal selective breeding programmes —however, they are currently not greatly outperforming more traditional approaches used to achieve this.

A review by an international team of scientists looked at several algorithms that have been used to analyse single nucleotide polymorphism (SNP) datasets, which feature variations at a single position in a DNA sequence among individuals of a species. Machine learning has been used to predict breeding values in dairy and beef cattle, pigs and broilers, they explained, as well as to predict the likelihood of disease occurrence.

Genomic prediction can be hampered by ever-increasing amounts of data, and the researchers found that machine learning models were effective in handling these large, complex datasets, and interpreting the interaction between genes.

However, in some cases, using them resulted in worse outcomes than when using conventional linear or mixed modelling approaches, which are generally associated with estimating the breeding value of traits that can be easily measured and have moderate to high heritability between generations.

Scope to iron out flaws

“The results of the reviewed studies showed that machine learning models have indeed performed well in fitting large noisy data sets and modelling minor nonadditive effects in some of the studies,” they explained in the journal Frontiers in Genetics.

“However, sometimes conventional methods outperformed machine learning models, which confirms that there’s no universal method for genomic prediction.”

Some of the flaws using currently available machine learning models include overfitting data to models and requiring very large ‘training’ datasets to be useful. They are also prone to identifying relationships between input and output variables which are not easy to understand.

The researchers stressed that the adoption of machine learning in genomic prediction remains in its infancy, recommending that further work needs to be done to explore their promise in a vast and fast-changing field.

Share this article...

You might also like...

Share this article...

Leave a Comment

Your email address will not be published. Required fields are marked *

Written by:

Sign up to our newsletter

FFF’s bi-weekly emails are filled with the latest news and information — sign up now to make sure the good stuff reaches your inbox. We promise we won’t send spam.
Subscription Form
Farming Future Food