Genome-wide association study and genomic prediction of Fusarium ear rot resistance in tropical maize germplasm

Fusarium ear rot(FER)is a destructive maize fungal disease worldwide.In this study,three tropical maize populations consisting of 874 inbred lines were used to perform genomewide association study(GWAS)and genomic prediction(GP)analyses of FER resistance.Broad phenotypic variation and high heritability for FER were observed,although it was highly influenced by large genotype-by-environment interactions.In the 874 inbred lines,GWAS with general linear model(GLM)identified 3034 single-nucleotide polymorphisms(SNPs)significantly associated with FER resistance at the P-value threshold of 1×10^(-5),the average phenotypic variation explained(PVE)by these associations was 3%with a range from 2.33%to 6.92%,and 49 of these associations had PVE values greater than 5%.The GWAS analysis with mixed linear model(MLM)identified 19 significantly associated SNPs at the P-value threshold of 1×10^(-4),the average PVE of these associations was 1.60%with a range from 1.39%to 2.04%.Within each of the three populations,the number of significantly associated SNPs identified by GLM and MLM ranged from 25 to 41,and from 5 to 22,respectively.Overlapping SNP associations across populations were rare.A few stable genomic regions conferring FER resistance were identified,which located in bins 3.04/05,7.02/04,9.00/01,9.04,9.06/07,and 10.03/04.The genomic regions in bins 9.00/01 and 9.04 are new.GP produced moderate accuracies with genome-wide markers,and relatively high accuracies with SNP associations detected from GWAS.Moderate prediction accuracies were observed when the training and validation sets were closely related.These results implied that FER resistance in maize is controlled by minor QTL with small effects,and highly influenced by the genetic background of the populations studied.Genomic selection(GS)by incorporating SNP associations detected from GWAS is a promising tool for improving FER resistance in maize.

Genomic prediction of the performance of hybrids and the combining abilities for line by tester trials in maize

The two most important activities in maize breeding are the development of inbred lines with high values of general combining ability(GCA)and specific combining ability(SCA),and the identification of hybrids with high yield potentials.Genomic selection(GS)is a promising genomic tool to perform selection on the untested breeding material based on the genomic estimated breeding values estimated from the genomic prediction(GP).In this study,GP analyses were carried out to estimate the performance of hybrids,GCA,and SCA for grain yield(GY)in three maize line-by-tester trials,where all the material was phenotyped in 10 to 11 multiple-location trials and genotyped with a mid-density molecular marker platform.Results showed that the prediction abilities for the performance of hybrids ranged from 0.59 to0.81 across all trials in the model including the additive effect of lines and testers.In the model including both additive and non-additive effects,the prediction abilities for the performance of hybrids were improved and ranged from 0.64 to 0.86 across all trials.The prediction abilities of the GCA for GY were low,ranging between-0.14 and 0.13 across all trials in the model including only inbred lines;the prediction abilities of the GCA for GY were improved and ranged from 0.49 to 0.55 across all trials in the model including both inbred lines and testers,while the prediction abilities of the SCA for GY were negative across all trials.The prediction abilities for GY between testers varied from-0.66 to 0.82;the performance of hybrids between testers is difficult to predict.GS offers the opportunity to predict the performance of new hybrids and the GCA of new inbred lines based on the molecular marker information,the total breeding cost could be reduced dramatically by phenotyping fewer multiple-location trials.

DNNGP, a deep neural network-based method for genomic prediction using multi-omics data in plants

Genomic prediction is an effective way to accelerate the rate of agronomic trait improvement in plants.Traditional methods typically use linear regression models with clear assumptions;such methods are unable to capture the complex relationships between genotypes and phenotypes.Non-linear models(e.g.,deep neural networks)have been proposed as a superior alternative to linear models because they can capture complex non-additive effects.Here we introduce a deep learning(DL)method,deep neural network genomic prediction(DNNGP),for integration of multi-omics data in plants.We trained DNNGP on four datasets and compared its performance with methods built with five classic models:genomic best linear unbiased prediction(GBLUP);two methods based on a machine learning(ML)framework,light gradient boosting machine(LightGBM)and support vector regression(SVR);and two methods based on a DL framework,deep learning genomic selection(DeepGS)and deep learning genome-wide association study(DLGWAS).DNNGP is novel in five ways.First,it can be applied to a variety of omics data to predict phenotypes.Second,the multilayered hierarchical structure of DNNGP dynamically learns features from raw data,avoiding overfitting and improving the convergence rate using a batch normalization layer and early stopping and rectified linear activation(rectified linear unit)functions.Third,when small datasets were used,DNNGP produced results that are competitive with results from the other five methods,showing greater prediction accuracy than the other methods when large-scale breeding data were used.Fourth,the computation time required by DNNGP was comparable with that of commonly used methods,up to 10 times faster than DeepGS.Fifth,hyperparameters can easily be batch tuned on a local machine.Compared with GBLUP,LightGBM,SVR,DeepGS and DLGWAS,DNNGP is superior to these existing widely used genomic selection(GS)methods.Moreover,DNNGP can generate robust assessments from diverse datasets,including omics data,and quickly incorporate complex and large datasets into usable models,making it a promising and practical approach for straightforward integration into existing GS platforms.