Variation analysis using random forests reveals domestication patterns and breeding trends in sugar beet

Cultivated beets (Beta vulgaris), including sugar beet, are important crops, and several studies employed whole genome sequencing to explore genomic variation. We applied the machine learning method "random forests" on hundreds of sequenced beet accessions and identified genomic variants that distinguish wild from domesticated beets at a mean accuracy of 98.4%. Associated genes were involved in sugar accumulation and transport (e.g., SUC4), nematode resistance, and root growth. Modern breeding lines from leading seed companies were distinguished from public seed bank accessions at 98.5% accuracy, revealing a strong signal linked to fungal resistance, likely originating from Italian wild beets. We also differentiated accessions by company, uncovering genes under selection, notably the flowering regulator APETALA1. Admixture profiles were analyzed to address open questions regarding the genomic history, provenance, and dispersal of wild beets. Our findings provide exciting possibilities for targeted breeding and show advances in variation analysis using machine learning.