Animals

The yak (Bos grunniens) thrives under chronic hypoxia and cold on the Qinghai–Tibet Plateau, yet a cross-tissue view of post-transcriptional regulation in this species remains limited. Here, we integrated multi-tissue RNA-seq and miRNA-seq data (tissues pooled from three Maiwa yaks) to construct and compare tissue-specific competing endogenous RNA (ceRNA) networks, while explicitly addressing a major source of false positives in ceRNA inference—misclassified lncRNA candidates with translational signatures. We cataloged 10,037 high-confidence lncRNAs (9360 non-redundant), 234 circRNAs, and 1030 miRNAs across six tissues. We then used Ribo-seq as an orthogonal quality-control layer to remove lncRNA candidates showing clear ribosome-association signals prior to network construction. Using a shared-target strategy (7mer-m8 seed matches; a ceRNA edge required ≥5 shared miRNAs), we assembled ceRNA networks for liver, lung, spleen, testis, and small intestine; skeletal muscle was excluded owing to insufficient Ribo-seq support for consistent filtering. Network topology varied substantially across tissues, with the testis network exhibiting the highest connectivity. ceRNA edges showed minimal overlap between tissues, indicating strong tissue dependence, whereas miRNA load/use profiles were moderately concordant, supporting a hierarchical conserved core—variable periphery organization. Importantly, the Ribo-seq–filtered lncRNA set provides a separate pool of ribosome-associated candidates for targeted follow-up, although ribosome association alone does not establish stable micropeptide production. Together, our results deliver a multi-tissue ceRNA resource and a reproducible, evidence-aware workflow for prioritizing candidate regulators while reducing annotation-driven false positives in yak.

Photoperiod and seasonality influence reproduction and lactation in sheep, but their effects on milk hormones, milk composition, and lamb growth are not fully understood. This study assessed the effect of season on milk prolactin, leptin, and insulin concentrations, milk chemical composition, lactation performance, and lamb growth in Polish Mountain ewes. Forty ewes were divided into the following two groups: short-day (lambing in December, n = 20) and long-day (lambing in May, n = 20). Milk samples were collected on days 5, 15, 25, 35, and 45 of lactation. Ewes in the long-day photoperiod had higher milk yield (p < 0.01) and higher prolactin and insulin concentrations (p < 0.01), whereas leptin concentrations did not differ seasonally. Milk from short-day ewes was characterized by higher dry matter and fat content (p < 0.01) and higher protein and lactose content (p < 0.05). Lambs from the long-day group achieved higher mean daily gain (p < 0.01). These results indicate that photoperiod influences lactation performance, milk composition, and offspring growth through seasonal hormonal and metabolic mechanisms, suggesting that appropriate lambing timing and day length manipulation can improve milk production efficiency and lamb growth in practical sheep production systems.

Artificial intelligence models, which have begun to be used in every field of science in recent years, have also started to come to the forefront in the classification of avians using bones. This study aimed to identify breeds of domestic fowl (Gallus gallus domesticus L. 1758) using morphometric measurements obtained from the tarsometatarsus bone and machine learning. A total of 328 tarsometatarsus specimens from two different modern domestic fowl breeds were used. A model was developed by performing 10 different morphometric measurements on each tarsometatarsus, and 3280 data points were obtained. Before model development, data cleaning and necessary assessments were carried out, and gaps were identified. In pre-processing and data partitioning, 70% of the data was used for training, and 30% was reserved for testing the developed model. To determine the differences between breeds, evaluations were performed using classical supervised learning algorithms in machine learning. Random forest (RF), support vector machine with radial kernel (SVM-RBF), and the generalized linear model (GLM, logistic regression) were used for model development, while model validation was performed using cross-validation (CV) metrics. After model validation, variable importance, feature selection, correlation analysis, dimensionality reduction, and multicollinearity were performed. The developed model, using morphological measurements obtained from the tarsometatarsus, distinguishes between breeds with high accuracy. The discriminative signal is extremely strong, allowing multiple modeling strategies (tree-based, kernel-based, and linear) to perfectly distinguish between the two breeds. Among the morphometric measurements, Ac (extension of the trochlea metatarsi IV) and Bmit (breadth of the middle trochlea) were found to be the strongest distinguishing features. This developed model combines morphometric data and artificial intelligence to offer an innovative method for scaling, accelerating, or improving applications in science. By expanding the model’s database with measurements obtained from the tarsometatarsus bones of different breeds, it was demonstrated that breed differences can be quickly and accurately determined using a minimal number of measurements from tarsometatarsus bones.