Predicting VNN resistance in European sea bass using machine learning on high dimensional low sample size data

Area of Science:

• Aquaculture
• Bioinformatics
• Genetics

Background:

• Aquaculture is a vital global food source, requiring accurate phenotype prediction for high-quality protein production.
• Predicting traits like disease resistance is crucial but challenging in aquaculture due to small sample sizes and high-dimensional genetic data (e.g., SNPs).
• This study addresses the common bioinformatics challenge of small sample, high-dimensional data in non-model species.

Purpose of the Study:

• To evaluate machine learning techniques for predicting Viral Nervous Necrosis (VNN) resistance in European sea bass.
• To assess the performance of various machine learning models under challenging data conditions, including genomic distance-based partitioning.
• To investigate the impact of dimensionality reduction using functional genetic information on prediction accuracy.

Main Methods:

• Explored Support Vector Machines, Gradient Boosting, Deep Learning, and Chaos Game Representation for phenotype prediction.
• Utilized a stringent dataset partitioning strategy maximizing genomic distance between training and testing sets.
• Reduced data dimensionality by selecting Single Nucleotide Polymorphisms (SNPs) based on functional information.

Main Results:

• The prediction of disease susceptibility in aquaculture remains a complex task.
• All tested machine learning tools demonstrated promising results in at least one evaluated scenario.
• Machine learning approaches, augmented with functional genetic data, show potential for mitigating challenges in high-dimensional, low-sample datasets.

Conclusions:

• Machine learning offers a promising avenue for phenotype prediction in aquaculture, particularly for non-model species.
• Integrating functional genetic information can enhance the effectiveness of machine learning models in aquaculture breeding programs.
• Despite inherent difficulties, machine learning can help address data scarcity issues in genetic studies within aquaculture.