Grain yield is increasingly affected by climate factors such as drought and heat. To develop resilient and high-yielding cultivars, high-throughput phenotyping (HTP) techniques are essential for precise decisions in wheat breeding. The ability of unmanned aerial vehicle (UAV)-based multispectral imaging and ensemble learning methods to increase the accuracy of grain yield prediction in practical breeding work is evaluated in this study. For this, 211 winter wheat genotypes were planted under full and limited irrigation treatments, and multispectral data were collected at heading, flowering, early grain filling (EGF), and mid-grain filling (MGF) stages. Twenty multispectral vegetation indices (VIs) were estimated, and VIs with heritability greater than 0.5 were selected to evaluate the models across the growth stages under both irrigation treatments. A framework for ensemble learning was developed by combining multiple base models such as random forest (RF), support vector machine (SVM), Gaussian process (GP), and ridge regression (RR). The R² values between VIs and grain yield for individual base models were ranged from 0.468 to 0.580 and 0.537 to 0.598 for grain yield prediction in full and limited irrigation treatments across growth stages, respectively. The prediction results of ensemble models were ranged from 0.491 to 0.616 and 0.560 to 0.616 under full and limited irrigation treatments respectively, and were higher than that of the corresponding base learners. Moreover, the grain yield prediction results were observed high at mid grain filling stage under both full (R² = 0.625) and limited (R² = 0.628) irrigation treatments through ensemble learning based stacking of four base learners. Further improvements in ensemble learning models can accelerate the use of UAV-based multispectral data for accurate predictions of complex traits like grain yield in wheat.

Increasing grain yield by the selection for optimal plant architecture has been the key focus in modern maize breeding. As a result, leaf angle, an important determinant of plant architecture, has been significantly improved to adapt to the ever-increasing plant density in maize production over the past several decades. To extend our understanding on the genetic mechanisms of leaf angle in maize, we developed the first four-way cross mapping population, consisting of 277 lines derived from four maize inbred lines with varied leaf angles. The four-way cross mapping population together with the four parental lines were evaluated for leaf angle in two environments. In this study, we reported linkage maps built in the population and quantitative trait loci (QTL) on leaf angle detected by inclusive composite interval mapping (ICIM). ICIM applies a two-step strategy to effectively separate the cofactor selection from the interval mapping, which controls the background additive and dominant effects at the same time. A total of 14 leaf angle QTL were identified, four of which were further validated in near-isogenic lines (NILs). Seven of the 14 leaf angle QTL were found to overlap with the published leaf angle QTL or genes, and the remaining QTL were unique to the four-way population. This study represents the first example of QTL mapping using a four-way cross population in maize, and demonstrates that the use of specially designed four-way cross is effective in uncovering the basis of complex and polygenetic trait like leaf angle in maize.

Genomic selection (GS) is a strategy to predict the genetic merits of individuals using genome-wide markers. However, GS prediction accuracy is affected by many factors, including missing rate and minor allele frequency (MAF) of genotypic data, GS models, trait features, etc. In this study, we used one wheat population to investigate prediction accuracies of various GS models on yield and yield-related traits from various quality control (QC) scenarios, missing genotype imputation, and genome-wide association studies (GWAS)-derived markers. Missing rate and MAF of single nucleotide polymorphism (SNP) markers were two major factors in QC. Five missing rate levels (0%, 20%, 40%, 60%, and 80%) and three MAF levels (0%, 5%, and 10%) were considered and the five-fold cross validation was used to estimate the prediction accuracy. The results indicated that a moderate missing rate level (20% to 40%) and MAF (5%) threshold provided better prediction accuracy. Under this QC scenario, prediction accuracies were further calculated for imputed and GWAS-derived markers. It was observed that the accuracies of the six traits were related to their heritability and genetic architecture, as well as the GS prediction model. Moore–Penrose generalized inverse (GenInv), ridge regression (RidgeReg), and random forest (RForest) resulted in higher prediction accuracies than other GS models across traits. Imputation of missing genotypic data had marginal effect on prediction accuracy, while GWAS-derived markers improved the prediction accuracy in most cases. These results demonstrate that QC on missing rate and MAF had positive impact on the predictability of GS models. We failed to identify one single combination of QC scenarios that could outperform the others for all traits and GS models. However, the balance between marker number and marker quality is important for the deployment of GS in wheat breeding. GWAS is able to select markers which are mostly related to traits, and therefore can be used to improve the prediction accuracy of GS.

Clonal species are common among plants. Clonal F ₁ progenies are derived from the hybridization between 2 heterozygous clones. In self- and cross-pollinated species, double crosses can be made from 4 inbred lines. A clonal F ₁ population can be viewed as a double cross population when the linkage phase is determined. The software package GACD (Genetic Analysis of Clonal F ₁ and Double cross) is freely available public software, capable of building high-density linkage maps and mapping quantitative trait loci (QTL) in clonal F ₁ and double cross populations. Three functionalities are integrated in GACD version 1.0: binning of redundant markers (BIN); linkage map construction (CDM); and QTL mapping (CDQ). Output of BIN can be directly used as input of CDM. After adding the phenotypic data, the output of CDM can be used as input of CDQ. Thus, GACD acts as a pipeline for genetic analysis. GACD and example datasets are freely available from www.isbreeding.net .

In this study, we considered four categories of molecular markers based on the number of distinguishable alleles at the marker locus and the number of distinguishable genotypes in clonal F1 progenies. For two marker loci, there are nine scenarios that allow the estimation of female, male, and/or combined recombination frequencies. In a double cross population derived from four inbred lines, five categories of markers are classified and another five scenarios are present for recombination frequency estimation. Theoretical frequencies of identifiable genotypes were given for each scenario, from which the maximum likelihood estimates of one or more of the three recombination frequencies could be estimated. If there was no analytic solution, then Newton-Raphson method was used to acquire a numerical solution. We then proposed to use an algorithm in Traveling Salesman Problem to determine the marker order. Finally, we proposed a procedure to build the two haploids of the female parent and the two haploids of the male parent in clonal F1. Once the four haploids were built, clonal F1 hybrids could be exactly regarded as a double cross population. Efficiency of the proposed methods was demonstrated in simulated clonal F1 populations and one actual maize double cross. Extensive comparisons with software JoinMap4.1, OneMap, and R/qtl show that the methodology proposed in this article can build more accurate linkage maps in less time.

Deep learning‐based genomic prediction (DL‐based GP) has shown promising performance compared to traditional GP methods in plant breeding, particularly in handling large, complex multi‐omics data sets. However, the effective development and widespread adoption of DL‐based GP still face substantial challenges, including the need for large, high‐quality data sets, inconsistencies in performance benchmarking, and the integration of environmental factors. Here, we summarize the key obstacles impeding the development of DL‐based GP models and propose future developing directions, such as modular approaches, data augmentation, and advanced attention mechanisms.