Nguyen Trung Duc

Why precision breeding

Edible maize (sweet, waxy and other specialty corn) is bred for eating quality as much as for yield. The genes behind that quality are mapped; progress now depends on the stage of breeding a program works at.

Starch and sugar (sweet, waxy)
Nutritional quality
Anthocyanin and pericarp color
Plant type, flowering and domestication
Disease resistance

Breeding 1.0

Farmer selection

Farmers kept sweet and glutinous mutants. Five known sugary1 alleles arose independently, two of them in the Andean landrace Chullpi and in Maíz Dulce from Mexico, and waxy maize was first found in China in 1908.

VNUA 160 local waxy maize accessions and 36 exotic sources assembled since 2003.

Breeding 2.0

Crosses, statistics and hybrids

Endosperm genes were bred into inbred lines and hybrids. The first supersweet (sh2) hybrid, Illini Xtra Sweet, was released in 1961, and sh2 is now in most hybrids for the fresh and processing markets.

VNUA Five co-developed hybrids recognized since 2022, chosen by combining ability and multi-trait indices. Like most programs in developing countries, the work is still mainly at this stage.

Breeding 3.0

Markers and genomic selection

Markers read the endosperm genes directly, and a 5,759-SNP array predicted kernel weight and size in 277 fresh waxy maize lines with accuracies of 0.33 to 0.64.

VNUA SSR markers linked to sh2 and su1, and a first doubled-haploid test with the inducer UH400.

Breeding 4.0

Genomes, editing, phenomics and machine learning

A sweet corn reference genome resolved the supersweet allele sh2-R, and CRISPR-edited waxy hybrids outyielded introgressed ones by 5.5 bushels per acre across 25 locations.

VNUA Image-based phenotyping and machine learning predicted rice nitrogen use efficiency with R² 0.98; maize is next.

Breeding 5.0

AI-centered breeding (proposed)

Proposed by several authors: foundation models, automated data collection, simulated breeding environments and explainable AI.

VNUA A breeding digital twin that predicts each selection on quality, yield, and nitrogen and phosphorus use efficiency, and audits it against later trials.

a, Start positions of 49 genes that define or improve edible maize on chromosomes 1–10 of the B73 reference genome (RefGen_v5, MaizeGDB gene models), drawn to one scale; color and marker shape give the trait category. Tightly linked genes that matter in breeding include a1–sh2 on chromosome 3, the reason purple supersweet corn is hard to breed. b, Stages of plant breeding, 1.0 to 4.0 after Wallace et al. (2018) and Aggarwal et al. (2026), with Breeding 5.0 as proposed; each card gives a milestone in edible maize and the position of the VNUA program. Plant silhouette: Onsemeliot, PhyloPic (public domain).

Sources: Hu et al. 2021, Nat. Commun.; Zheng et al. 2013, PLoS ONE; Tracy et al. 2019, Plant Breed. Rev.; Anirban et al. 2023, Sci. Rep.; Qu et al. 2026, Plants; Gao et al. 2020, Nat. Biotechnol.; Wallace et al. 2018, Annu. Rev. Genet.; Aggarwal et al. 2026, Theor. Appl. Genet.; MaizeGDB, B73 RefGen_v5.