Genomic prediction has emerged as a powerful tool in aquaculture breeding, but its effectiveness depends on the careful selection of informative single nucleotide polymorphisms (SNPs) and the application of appropriate prediction models. This study aimed to enhance genomic prediction accuracy in Russian sturgeon (Acipenser gueldenstaedtii) by optimizing SNP selection strategies and exploring the performance of linear and machine learning models. Three economically important traits—caviar yield, caviar color, and body weight—were selected due to their direct relevance to breeding goals and market value. Whole-genome sequencing (WGS) data were obtained from 971 individuals with an average sequencing depth of 13.52×. To reduce marker density and eliminate redundancy, three SNP selection strategies were applied: (1) genome-wide association study (GWAS)-based prioritization to select trait-associated SNPs; (2) linkage disequilibrium (LD) pruning to retain independent markers; and (3) random sampling as a control. Genomic prediction was conducted using both linear (e.g., GBLUP) and machine learning models (e.g., random forest) across varying SNP densities (1 K to 50 K). Results showed that GWAS-based SNP selection consistently outperformed other strategies, especially at moderate densities (≥10 K), improving prediction accuracy by up to 3.4% compared to the full WGS dataset. LD-based selection at higher densities (30 K and 50 K) achieved comparable performance to full WGS. Notably, machine learning models, particularly random forest, exceeded the performance of linear models, yielding an additional 2.0% increase in accuracy when combined with GWAS-selected SNPs. In conclusion, integrating WGS data with GWAS-informed SNP selection and advanced machine learning models offers a promising framework for improving genomic prediction in sturgeon and holds promise for broader applications in aquaculture breeding programs.

Caviar yield, caviar color, and body weight are crucial economic traits in sturgeon breeding. Understanding the molecular mechanisms behind these traits is essential for their genetic improvement. In this study, we performed whole-genome sequencing on 673 Russian sturgeons, renowned for their high-quality caviar. With an average sequencing depth of 13.69×, we obtained approximately 10.41 million high-quality single nucleotide polymorphisms (SNPs). Using a genome-wide association study (GWAS) with a single-marker regression model, we identified SNPs and genes associated with these traits. Our findings revealed several candidate genes for each trait: caviar yield: TFAP2A, RPS6KA3, CRB3, TUBB, H2AFX, morc3, BAG1, RANBP2, PLA2G1B, and NYAP1; caviar color: NFX1, OTULIN, SRFBP1, PLEK, INHBA, and NARS; body weight: ACVR1, HTR4, fmnl2, INSIG2, GPD2, ACVR1C, TANC1, KCNH7, SLC16A13, XKR4, GALR2, RPL39, ACVR2A, ADCY10, and ZEB2. Additionally, using the genomic feature BLUP (GFBLUP) method, which combines linkage disequilibrium (LD) pruning markers with GWAS prior information, we improved genomic prediction accuracy by 2%, 1.9%, and 3.1% for caviar yield, caviar color, and body weight traits, respectively, compared to the GBLUP method. In conclusion, this study enhances our understanding of the genetic mechanisms underlying caviar yield, caviar color, and body weight traits in sturgeons, providing opportunities for genetic improvement of these traits through genomic selection.

Genomic prediction using whole-genome sequencing (WGS) data is challenged by the imbalance between a limited sample size (n) and an extensive number of single-nucleotide polymorphisms (SNPs) (p), where n $\ll$p. The high dimensionality of WGS data also increases computational demands, limiting its practical application. In this study, we introduce DAGP, a novel method that integrates deep autoencoder compression to reduce WGS data dimensionality by over 99% while preserving essential genetic information. This compression significantly improves computational efficiency, facilitating the effective use of high-dimensional genomic data. Our results demonstrated that DAGP, when combined with the genomic best linear unbiased prediction (GBLUP) method, maintained prediction accuracy comparable to WGS data, even at reduced marker densities of 50 K for sturgeon and 20 K for maize. Furthermore, integrating DAGP with Bayesian and machine learning models improved genomic prediction accuracy over traditional WGS-based GBLUP, with an average gain of 6.05% and 5.35%, respectively. DAGP provides an efficient and scalable solution for genomic prediction in species with large-scale genomic data, offering both computational feasibility and enhanced prediction performance.

Masu salmon (Oncorhynchus masou) is an ecologically and economically important cold-water salmonid in East Asia that exhibits diverse life-history forms. To compare population-level genomic variation and patterns of homozygosity among fish from different sources, we analyzed whole-genome resequencing data from 465 individuals representing one field-collected Tumen River population (TM) and two landlocked cultured populations from Chicheng (CC) and Yanji (YJ). After quality control, 6,220,980 high-quality SNPs were retained. Population-specific filtering identified 5,589,828, 3,545,209, and 4,975,751 polymorphic SNPs in CC, TM, and YJ, respectively; although SNP numbers differed, approximately 91% of variants in each population were located in intronic or intergenic regions. Principal component analysis, ADMIXTURE, and distance-based neighbor-joining analysis clearly distinguished the three populations, with CC and YJ showing the closest genetic relationship. Pairwise $F_{\mathrm{ST}}$ was lowest between CC and YJ and highest between TM and YJ. CC exhibited the highest linkage disequilibrium, whereas TM showed the fastest LD decay and the lowest nucleotide diversity and heterozygosity. Runs of homozygosity (ROH) burden was highest in TM, intermediate in CC, and lowest in YJ. TM had the highest number of ROHs, cumulative ROH length, and $F_{\mathrm{ROH}}$, and ROHs longer than 5 Mb were detected only in this population. CC had an intermediate ROH burden dominated by short segments, whereas YJ had the lowest ROH-based genomic inbreeding. The high and heterogeneous ROH burden in TM indicates elevated genome-wide homozygosity among the sampled fish but does not, by itself, demonstrate recent inbreeding throughout the population. Candidate ROH islands and their annotated genes showed limited overlap among populations. Candidate genes in TM were primarily associated with ion regulation, neural processes, and energy metabolism, whereas those in CC and YJ shared broad functional categories involving development, muscle organization, nutrient transport, and neural regulation but differed in most specific genes. These regions and genes should be regarded as exploratory, hypothesis-generating candidates rather than evidence of selection or causality. Overall, this study reveals distinct population genomic characteristics and ROH patterns among masu salmon populations of different origins and provides a basis for future germplasm conservation and genetic management.

Accurate pedigree reconstruction is essential for genetic evaluation, inbreeding control, and family management in aquaculture breeding programs. In sturgeon, extremely high fecundity and communal rearing during early developmental stages often lead to the loss of family information, making reliable full-sib family assignment a critical challenge. In this study, we developed a machine learning-based framework for full-sib family assignment using simulated genotype datasets and whole-genome resequencing data from Russian sturgeon. Simulation analyses across five machine learning algorithms showed that training set size and marker density were the primary determinants of assignment accuracy. When at least 10 individuals per family were included in the training set, mean identification accuracy exceeded 99% across all evaluated scenarios, and exceeded 99.9% for all methods except XGBoost. In contrast, performance declined when the marker number was reduced to 200. At moderate marker densities (500–1000 markers), performance remained stable, with mean identification accuracy around 99% even when only 3–4 individuals per family were included in the training set. Validation using whole-genome resequencing data (sequencing depth ranging from 9.43× to 11.86×) from 582 individuals representing 19 full-sib families of Russian sturgeon confirmed the simulation findings, with several algorithms achieving assignment accuracies exceeding 99%. These results demonstrate that machine learning provides an accurate and robust approach for full-sib family assignment using genome-wide single nucleotide polymorphism (SNP) data. The proposed framework offers an effective solution for pedigree reconstruction and family identification in sturgeon breeding populations lacking reliable pedigree records.