Revolutionizing solanaceae breeding: Integrating genomic selection and machine learning for enhanced trait predictability
image: Fig.1 GS predictabilities in tomato and pepper.
Credit: Horticulture Research
Plant breeding is critical to the advancement of horticultural, especially in the Solanaceae family (including tomato, pepper, and eggplant), which relies on selecting high-performance lines for desired traits. Traditional breeding methods, though effective, face challenges in precision and labor intensity. The advent of machine learning and genomic prediction, particularly Genomic Selection (GS), harnesses genetic markers like SNPs to estimate breeding values of unseen lines, promising a revolution in breeding efficiency. Current research focuses on overcoming limitations of marker-assisted selection (MAS) by integrating various GS models (regression-based, classification-based, and deep learning) to enhance predictability across diverse traits. The primary challenge lies in refining these models to accurately predict minor gene effects and streamline breeding processes.
In March 2022, Horticulture Research published a research article entitled by “Genomic prediction of morphometric and colorimetric traits in Solanaceous fruits”.
This study investigates the performance of genomic selection (GS) models for morphometric and colorimetric traits in Solanaceous fruits, specifically tomatoes and peppers.These traits were measured by score-based traditional descriptors (CD) as well as by the Tomato Analyser (TA) tool using longitudinal and transverse cut fruit images. TA traits and marker SNPs were a powerful combination for predicting morphological and colour-related traits in Solanaceae fruits The study revealed that pepper showed higher predictability than tomato across all TA traits. Intriguingly, basic traits like fruit size were more easily predicted from genomic information than more complex traits. For certain chromaticity traits in peppers, multi-trait GS models slightly outperformed single-trait models. Testing on an independent population of wild tomatoes revealed lower predictability for all TA traits, likely due to the genetic distance between the training (cultivated) and testing (wild) populations. Adding just a few wild accessions dramatically improved the predictability, underscoring the importance of diverse germplasm in GS. For traits scored by CDs, classification-based GS model was compared with the TA phenotyping. The predictability of CD traits was generally lower than related TA traits, suggesting that the TA phenotype was superior in predicting fruit size and yield-related traits.
In conclusion, the study demonstrates varied predictabilities across different GS models and traits in Solanaceous fruits. It highlights the impact of genetic diversity on model performance and underscores the efficiency of TA over conventional phenotyping methods. The findings provide valuable insights into the optimization of GS models for improved plant breeding outcomes.
###
References
Authors
Hao Tong1,2,3, Amol N. Nankar1, Jintao Liu2, Velichka Todorova4, Daniela Ganeva4, Stanislava Grozeva4, Ivanka Tringovska4, Gancho Pasev4, Vesela Radeva-Ivanova4, Tsanko Gechev1, Dimitrina Kostova1,4 and Zoran Nikoloski1,2,3,*
Affiliations
1Center of Plant Systems Biology and Biotechnology, Plovdiv, 4000, Bulgaria
2Systems Biology and Mathematical Modeling, Max Planck Institute of Molecular Plant Physiology, Potsdam, 14476, Germany
3Bioinformatics, Institute of Biochemistry and Biology, University of Potsdam, Potsdam, 14476, Germany
4Maritsa Vegetable Crops Research Institute, Plovdiv, 4003, Bulgaria.
About Zoran Nikoloski
Zoran Nikoloski is Professor at the University of Potsdam and group leader at the Max Planck Institute of Molecular Plant Physiology, Potsdam-Golm. Nikoloski’s early work focused on the development of computational approaches for integration of heterogeneous data from high-throughput molecular profiling technologies. Nikoloski has developed extensive expertise in computational systems biology with the aim of reconstructing networks from molecular profiles and understanding how network structure determined abundance of network components. Nikoloski’s main research areas include: (1) data-driven qualitative and quantitative modelling of genome-scale metabolic and gene-regulatory networks, (2) analysis of evolutionary and optimisation processes in biological networks, and (3) characterization of system’s functions emerging from molecular interactions.