Machine learning approaches using freshwater gene expression profiles to predict seawater performance in Atlantic Salmon

• Main Author: Robertsen, Sofie
• Other Authors: Simen Rød Sandve, Torgeir Rhoden Hvidsten
• Format: Master Thesis
• Language: English
• Published: Norwegian University of Life Sciences 2023
• Subjects: Atlantic salmon; Salmo salar
• Online Access: https://hdl.handle.net/11250/3076849

Atlantic salmon (Salmo salar) has an anadromous life cycle, spending the first part of its life in freshwater before migrating to seawater. Smoltification is the process where Atlantic salmon undergo several morphological, physiological and behavioral changes preparing for transition to marine environment. A major challenge in the Norwegian salmon farming industry is the high mortality (12-14%), after release of smolt into seawater. One reason is suboptimal smolt production, resulting in a state where salmon are not well adopted for life in seawater. It is therefore important to optimize smolt production protocols and develop better ways to assess seawater-readiness to ensure higher survival, growth and reduce welfare issues. Traditionally, the increased expression of the saltwater isoform nkaα1b and nkcc1a cotransporter, and a reduction in expression of the freshwater isoform nkaα1b in the gills are used as predictive markers for seawater-readiness in the salmon farming industry. The current study aimed to use Random Forest to build predictive models for growth in seawater based on gill transcriptome data from fish given different light manipulation during smolt production. The results showed poor predictive ability towards seawater growth, although superior to simple correlation with single gene expression levels. We also found that photoperiodic history had effect on the Random Forest predictions, where the Random Forest model from fish exposed to continuous light (24:0) was much better at predicting SW growth than any of the models from the fish exposed to short photoperiods (8:16 and 12:12). We extracted most influential genes for each Random Forest model and found that these differed depending on the light regime used. Based on these results the salmon farming industry should apply caution when relying on traditional smolt gene-expression markers to determine the optimal time for SW transfer.