AutoGP: An intelligent breeding platform for enhancing maize genomic selection.

Area of Science:

• Agricultural Science
• Genetics
• Computational Biology

Background:

• Climate change and population growth necessitate accelerated development of high-yielding crop varieties.
• Existing genotype-to-phenotype (G2P) models are hindered by high single-nucleotide polymorphism (SNP) density requirements and laborious phenotypic data collection.
• Advanced genomic selection (GS) models, like deep learning (DL), are often complex for breeders to implement.

Purpose of the Study:

• To develop an intelligent breeding platform, AutoGP, integrating genotype extraction, phenotype extraction, and GS models.
• To provide a user-friendly web interface for breeders to streamline G2P data analysis and genomic selection.
• To overcome limitations of existing G2P models by improving SNP identification and phenotype data acquisition.

Main Methods:

• Developed an efficient sequencing chip for high-quality SNP identification within gene-regulatory networks.
• Created a workflow for extracting plant phenotypes (e.g., plant height, leaf count) from smartphone videos.
• Integrated a diverse pool of five machine learning (ML) and four deep learning (DL) models for genomic prediction.

Main Results:

• AutoGP's genotype chips effectively extracted high-quality SNPs linked to traits like days to tasseling and plant height in maize.
• The integrated ML and DL models demonstrated reliable predictive accuracy across different populations.
• The platform successfully guided breeders using maize datasets, showcasing its practical utility.

Conclusions:

• AutoGP offers a practical and user-friendly solution for breeders to enhance genomic selection efficiency and accuracy.
• The platform streamlines the integration of genotype and phenotype data, facilitating faster crop improvement.
• AutoGP empowers breeders with advanced computational tools, addressing critical needs in modern agriculture.