Sex identification in rainbow trout using genomic information and machine learning

Area of Science:

• Aquaculture
• Genomics
• Machine Learning

Background:

• Accurate sex identification in farmed fish is vital for effective stock management and breeding programs, especially in juvenile stages where visual determination is challenging.
• The increasing availability of genomic data in aquaculture, coupled with diverse sex determination systems in ray-finned fishes, necessitates advanced identification methods.
• Existing genomic sex identification methods often rely on pre-identified markers and probabilistic approaches, which may not be universally applicable across populations.

Purpose of the Study:

• To develop and evaluate a novel machine learning approach for predicting sex in farmed fish using genomic data.
• To assess the performance of the Extreme Gradient Boosting (XGBoost) method on both simulated and real datasets, particularly when the suitability of genetic markers is unknown a priori.
• To determine the accuracy and robustness of the proposed method across varying genotyping error rates.

Main Methods:

• Utilized the Extreme Gradient Boosting (XGBoost) algorithm, a supervised machine learning technique, for sex prediction from unimputed genomic data.
• Assessed the method's accuracy using four simulated datasets with controlled genotyping error rates (5% and 50%) and one real dataset from the Finnish Rainbow Trout Breeding Program.
• Compared the prediction accuracy of the XGBoost approach against established methods where marker suitability is known beforehand.

Main Results:

• The XGBoost method demonstrated high prediction quality on both simulated and real datasets.
• Achieved perfect accuracy (1.0) on simulated data with a 5% genotyping error rate and 0.60 accuracy with a 50% error rate.
• In the real rainbow trout dataset, the method attained a prediction accuracy of 98%, indicating its suitability for practical, routine application in aquaculture.

Conclusions:

• Supervised machine learning, specifically Extreme Gradient Boosting, offers a powerful and accurate tool for sex identification in farmed fish using genomic data, even without prior knowledge of sex-linked markers.
• The developed method is robust and performs well across different genotyping error levels, making it a valuable asset for aquaculture breeding programs.
• The high accuracy achieved in real-world data suggests this approach can be readily implemented for efficient and reliable sex determination in routine aquaculture practices.