Gene Expression-Based Classification of European Seabass Larval Batches According to Saddleback Syndrome Incidence Using Machine Learning

Skeletal deformities remain a major challenge in marine fish hatcheries, affecting larval quality, animal welfare, production efficiency, and market value. In European seabass (Dicentrarchus labrax), saddleback syndrome (SBS) is a relevant skeletal abnormality that develops during larval ontogeny and has been associated with defects of the primordial marginal finfold around the flexion stage. This study investigated whether gene expression markers, combined with machine learning, could provide a stage-specific molecular approach for assessing SBS-associated larval batch quality. Larval populations from commercial hatcheries were classified as GOOD or POOR according to SBS incidence at mid-metamorphosis. Gene expression was analyzed at first feeding, flexion, post-flexion, and mid-metamorphosis, targeting genes involved in osteogenesis, myogenesis, metabolism, oxidative phosphorylation, and stress response. Stage-specific random forest models were used to classify gene expression profiles derived from larval populations with contrasting SBS incidence and to identify candidate informative genes. The models showed cross-validated area under the receiver operating characteristic curve (ROC AUC) values ranging from 0.83 to 0.962, with the highest performance at flexion. Reduced models based on the three most informative genes retained comparable internal cross-validation performance. Key candidate genes were mainly related to mitochondrial energy production, iron metabolism, stress response, muscle development, and extracellular matrix formation. These findings suggest that gene expression profiling combined with machine learning may support stage-aware discrimination of larval populations with contrasting SBS incidence, although validation in larger independent datasets is required before hatchery application.