A Machine Learning Framework for Concurrent Identification of Favorable, Deleterious, and Lethal Variants in Bovine Populations

Abstract

Background. Accurate genomic screening in livestock is essential for managing a complex spectrum of traits, including favorable production-related alleles, deleterious variants, and recessive lethal alleles. However, most organizations rely on legacy analytic pipelines that tend to be computationally slow and which suffer from high technical debt. These legacy systems are often difficult to maintain, lack modularity, and create significant bottlenecks as marker panels expand.

Methods. We developed a high-throughput machine learning framework for identifying carrier status for many different haplotypes which will replace a legacy SAS program. By transitioning to a modular, Python-based architecture utilizing XGBoost and Polars, we streamlined the codebase, significantly reducing maintenance overhead and improving system extensibility. The new framework uses an automated feature-engineering pipeline to process genotype data, allowing for the concurrent classification of multiple favorable, sub-lethal deleterious, and recessive lethal variants in a single, reproducible environment.

Results. The ML-based approach demonstrated robust performance across all haplotypes tested, achieving an average concordance of 99.8% relative to the legacy system. In addition to matching the accuracy of the legacy system, the framework produced comparable runtimes on traditional CPU hardware and significantly reduced runtimes on modest GPU hardware. Crucially, the modernization of the codebase resulted in a 50% reduction in lines of code, enabling markedly faster updates as new genetic markers are discovered and integrated into the screening panel.

Conclusion. Modernizing genomic screening with machine learning provides high-performance trait identification and a sustainable, maintainable engineering foundation with minimal drawbacks. By reducing technical debt and increasing throughput, this framework allows genomic labs to pivot quickly to new research requirements while maintaining rigorous operational standards for population health and genetic progress.