The Hutton Potato Team collaborates closely with breeding partners to improve genetic gains by leveraging the groundbreaking discovery of the S-locus inhibitor gene (Sli-gene) and the unparalleled capabilities of AI technologies, particularly Machine Learning (ML).

Our roadmap encompasses germplasm exploration and development, genetic studies and pre-breeding at diploid levels, and machine learning-informed breeding at tetraploid levels. Our ultimate goal is to establish a comprehensive system that effectively unlocks the potential of traits within the Common Potato Collection (CPC) and seamlessly integrates them into downstream procedures for genetic studies and breeding purposes.

Key Objectives of our strategy:

  • Germplasm Exploration and Development:We are converting CPC germplasms into Recombinant Inbred Lines (RILs) by crossing with Sli-gene donors and subsequent selfing. These RIL collections, replacing the original highly heterozygous accessions, will provide geneticists and breeders with novel and more manageable germplasm resources. This streamlined approach will enhance the utilization of valuable traits within the CPC.
  • Development of Specific Populations for Genetic Studies and Pre-Breeding:To study novel traits, we are creating various genetic populations, including F2 populations, and, based on Sli, RILs, BILs and NILs. Simultaneously, we will develop inbred lines from induced Dihaploids sourced from adapted tetraploid cultivars with known disease-resistance genes, as well as from diploid cultivars such as Mayan Gold and elite lines from the experimental diploid population 06H1. Using this multipronged approach we aim to generate a series of inbred lines with desirable agronomic traits, suitable for hybrid breeding.
  • Machine Learning-Informed Breeding: Our strategy capitalises on BrenSeq (Breeding-oriented enrichment sequencing), our state-of-the-art sequencing platform tailored specifically for breeding purposes. With approximately 600 functional genes compiled, we have generated BrenSeq data for a diverse breeding panel encompassing different traits. Our association analysis has identified numerous highly significant candidate genes, reaffirming their role in trait expression. Furthermore, we have harnessed the power of Machine Learning (ML) technology for trait modeling and prediction, using algorithms like Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Kernel Density Estimation (KDE). The integration of ML has significantly enhanced prediction accuracy by over 10%, and we continue to refine our ML models through expanded phenotypic and BrenSeq datasets in collaboration with our industrial partners.