Title: PYTHON BASED GENETIC EVALUATION SYSTEM FOR THE IMPROVEMENT OF MULTI-BREED AND CROSSBRED BEEF CATTLE

Abstract:

The beef cattle industry faces the challenge of balancing sustainability and profitability. Technological advancements offer tools to address this challenge. AgSights provides a commercial, user-friendly genetic evaluation system (GES). However, its current implementation in lower-level programming languages (C and Fortran) limits its adaptability. In contrast, Python is a high-level programming language designed for developer productivity. Its adoption could facilitate rapid integration of new functionalities and easier maintenance over time. This thesis investigated the feasibility of using Python to update AgSights' GES. The Python GES was developed in-house and encompasses functionalities for data cleaning, renumbering, and formatting. It can calculate relationship coefficients and related matrices, including inbreeding coefficients and the inverse of the pedigree or hybrid (pedigree + genomic) relationship matrix. The software employs a preconditioned conjugate gradient algorithm with iteration on data to predict breeding values. In addition to re-implementing AgSights’ current pedigree-based GES, the software also facilitates the incorporation of genomic information through single-step genomic best linear unbiased prediction (ssGBLUP) to modernize the GES. Pedigree BLUP (Study 1) and ssGBLUP (Study 2) were applied to multiple trait models using simulated datasets representative of a beef breeding program for growth traits. In Study 1, the solver component for the prediction of over 2.9 million random effects took 12 minutes and 50 seconds. The addition of up to 10,000 genotyped animals to the evaluation increased computing time to 39 minutes and 28 seconds. Additionally, pedigree BLUP was tested using real producer-collected data encompassing a crossbred and multiple-breed population (Studies 3 and 4). Study 3 employed a single-trait model, evaluating 171,606 animals in approximately 2 minutes. Study 4 used a multiple-trait maternal model, evaluating over 3.2 million animals in just over 13 hours. Across all scenarios, the entire genetic evaluation process was completed within a day, maintaining accurate EBV estimates. Overall, it can be concluded that Python provides a suitable framework for developing flexible GES for modern beef cattle and, more generally, livestock breeding programs. The shift towards Python provides an opportunity for more innovative and efficient genetic evaluation practices, ultimately promoting a more sustainable and profitable livestock industry.