AI-Guided Marker Selection for Low-Cost Breeding Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for plant breeding, such as marker-assisted selection (MAS) and genomic prediction (GP), are limited by high costs, low efficiency, and population specificity, making it difficult to select progeny with desired phenotypic characteristics due to the need for extensive genotyping and costly production of QTL mapping populations.
Innovation Solution
A method using artificial intelligence and machine learning to simulate progeny populations, identify quantitative trait loci (QTLs), and select polymorphic markers associated with desired traits, reducing the number of genetic markers needed for genomic estimated breeding values (GEBVs) through a machine learning model trained on genotypic information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If genomic prediction (GP) with moderate density genotyping is used to estimate Genomic Breeding Values (GEBVs) for directional selection, then selection accuracy is improved, but genotyping costs become financially unfeasible due to the enormous numbers of generated embryos
Solution Approach 1:
The invention extracts and focuses on only the most informative genetic markers (QTLs) that have the greatest impact on trait prediction, rather than using moderate density genotyping across the entire genome. This extraction of critical information points reduces genotyping costs while maintaining selection accuracy, directly resolving the contradiction between measurement precision and quantity of substance.
Solution Approach 2:
The genome is segmented into functional units (QTLs) that are specifically associated with traits of interest. By dividing the genomic information into these functional segments and only genotyping markers within or near these QTL regions, the method reduces the overall genotyping burden while preserving the essential information needed for accurate selection, thereby addressing the cost-accuracy tradeoff.
2Quantity of substance
If marker-assisted selection (MAS) with a few well characterized QTLs is used to reduce genotyping costs, then genotyping costs are reduced, but performance is lower compared to GP and QTLs are population-specific
Solution Approach 1:
The invention performs preliminary action by using in silico crossing and machine learning models to predict phenotypes and identify QTLs before actual genotyping occurs. This preliminary computational phase allows the method to pinpoint the most relevant QTLs for each specific breeding population, ensuring that subsequent genotyping focuses on population-specific markers that maximize prediction accuracy while minimizing costs, thus overcoming the limitations of both traditional MAS and GP.
Solution Approach 2:
The method dynamically changes the parameters of marker selection based on the specific population being bred. By using machine learning to identify population-specific QTLs and adjusting which markers are genotyped accordingly, the system adapts to different germplasm populations, maintaining high prediction performance across diverse populations while keeping genotyping costs low, thereby resolving the population-specific limitation of traditional MAS.
3Loss of information
If QTL mapping populations are produced to identify markers for MAS, then markers can be identified, but the production of QTL mapping populations is costly and requires generating experimental populations in the field
Solution Approach 1:
The invention creates virtual copies of breeding populations through in silico crossing simulations. Instead of physically producing QTL mapping populations in the field, the method uses computational models to simulate crosses and predict offspring genotypes and phenotypes. This digital copying approach identifies QTLs and associated markers without the costly and time-consuming process of generating and phenotyping actual experimental populations, directly resolving the contradiction between marker identification and production cost.
Solution Approach 2:
The method replaces the mechanical/biological system of physical QTL mapping population production with a computational system. Machine learning models and in silico simulations substitute for field-based experimental population generation, allowing QTL identification and marker discovery to occur through computational analysis rather than physical breeding and phenotyping, thereby eliminating the high costs and logistical requirements of traditional QTL mapping.
Data Source
AI summary
The present disclosure provides improved breeding methods that allow for the selection of a member or members of a population having a desired phenotype using a limited number of genetic markers. In certain aspects, the methods for selecting members of the breeding program utilize machine learning models to predict phenotypes of a simulated progeny population.