Machine Learning-Based Accurate Full-Sib Family Assignment in Sturgeon Using Whole-Genome Sequencing Data.

Area of Science:

• Aquaculture Genetics
• Bioinformatics
• Machine Learning Applications

Background:

• Accurate pedigree reconstruction is vital for genetic improvement and managing aquaculture breeding programs.
• Sturgeon breeding faces challenges in family identification due to high fecundity and communal rearing, leading to lost pedigree information.

Purpose of the Study:

• To develop and validate a machine learning framework for accurate full-sib family assignment in Russian sturgeon.
• To assess the impact of training set size and marker density on family assignment accuracy.

Main Methods:

• Developed a machine learning framework utilizing simulated and whole-genome resequencing data.
• Evaluated five machine learning algorithms for full-sib family assignment.
• Tested performance across various training set sizes and single nucleotide polymorphism (SNP) marker densities.

Main Results:

• High assignment accuracy (exceeding 99%) was achieved with at least 10 individuals per family in the training set.
• Marker density significantly impacts accuracy; moderate densities (500-1000 SNPs) maintained high accuracy with fewer individuals.
• Validation on real Russian sturgeon data confirmed the framework's robustness, with accuracies over 99%.

Conclusions:

• Machine learning offers a powerful and reliable method for full-sib family assignment in sturgeon using genomic data.
• The developed framework provides an effective solution for pedigree reconstruction in aquaculture populations with incomplete records.