The validation and application of an ovine fertility model using standardised in vitro thresholds to predict the likelihood of pregnancy.

Deciphering a ram or ewe's reproductive potential is crucial to ensure high reproductive performance and maximise production outcomes. This study validates the accuracy of an ovine fertility model created to predict the likelihood of pregnancy occurring following laparoscopic artificial insemination (AI) and proposes in vitro semen standards to improve pregnancy outcomes. Semen from Merino sires (N = 26) was inseminated into synchronised Merino ewes (N = 1269) across 3 breeding seasons (2021-2023). Uterine tone and intra-abdominal fat of ewes were scored at AI, while the freezing concentration, abnormal sperm, acrosome viability (6h) and CASA motility and velocity traits (0h) of semen inseminated was assessed post-thaw (6h; 37 °C). Pregnancy predictions were compared with ultrasound-confirmed pregnancies ∼55 days post-AI, using discrimination and calibration tests to correctly assess its ability to classify pregnant and non-pregnant ewes. The model demonstrated high accuracy (77 %), precision (96 %) and recall (76 %) but lower specificity (33 %). It recorded an F1-score of 0.85, with an Area Under the Curve (AUC) of 0.62. There was no statistical difference between predicted and actual pregnancy results (P = 0.184) despite an error value of 26 %. A cutting point split the data for each in vitro semen predictor and calculated the average pregnancy rate above and below this point. The cutting point with the greatest difference between pregnancy rates was chosen as the semen threshold. When entered into the model, these thresholds returned a cumulative pregnancy probability of 64.3 %. These standards could be used to screen semen before AI, reducing the variability of laparoscopic AI programs for the industry.