Genomic prediction for rust resistance in pea.

Genomic selection (GS) has become an indispensable tool in modern plant breeding, particularly for complex traits. This study aimed to assess the efficacy of GS in predicting rust (Uromyces pisi) resistance in pea (Pisum sativum), using a panel of 320 pea accessions and a set of 26,045 Silico-Diversity Arrays Technology (Silico-DArT) markers. We compared the prediction abilities of different GS models and explored the impact of incorporating marker × environment (M×E) interaction as a covariate in the GBLUP (genomic best linear unbiased prediction) model. The analysis included phenotyping data from both field and controlled conditions. We assessed the predictive accuracies of different cross-validation strategies and compared the efficiency of using single traits versus a multi-trait index, based on factor analysis and ideotype-design (FAI-BLUP), which combines traits from controlled conditions. The GBLUP model, particularly when modified to include M×E interactions, consistently outperformed other models, demonstrating its suitability for traits affected by complex genotype-environment interactions (GEI). The best predictive ability (0.635) was achieved using the FAI-BLUP approach within the Bayesian Lasso (BL) model. The inclusion of M×E interactions significantly enhanced prediction accuracy across diverse environments in GBLUP models, although it did not markedly improve predictions for non-phenotyped lines. These findings underscore the variability of predictive abilities due to GEI and the effectiveness of multi-trait approaches in addressing complex traits. Overall, our study illustrates the potential of GS, especially when employing a multi-trait index like FAI-BLUP and accounting for M×E interactions, in pea breeding programs focused on rust resistance.

Biologically meaningful genome interpretation models to address data underdetermination for the leaf and seed ionome prediction in Arabidopsis thaliana.

Genome interpretation (GI) encompasses the computational attempts to model the relationship between genotype and phenotype with the goal of understanding how the first leads to the second. While traditional approaches have focused on sub-problems such as predicting the effect of single nucleotide variants or finding genetic associations, recent advances in neural networks (NNs) have made it possible to develop end-to-end GI models that take genomic data as input and predict phenotypes as output. However, technical and modeling issues still need to be fixed for these models to be effective, including the widespread underdetermination of genomic datasets, making them unsuitable for training large, overfitting-prone, NNs. Here we propose novel GI models to address this issue, exploring the use of two types of transfer learning approaches and proposing a novel Biologically Meaningful Sparse NN layer specifically designed for end-to-end GI. Our models predict the leaf and seed ionome in A.thaliana, obtaining comparable results to our previous over-parameterized model while reducing the number of parameters by 8.8 folds. We also investigate how the effect of population stratification influences the evaluation of the performances, highlighting how it leads to (1) an instance of the Simpson's Paradox, and (2) model generalization limitations.

The Effect of Genome Parametrization and SNP Marker Subsetting on Genomic Selection in Autotetraploid Alfalfa.

BACKGROUND: Alfalfa, the most economically important forage legume worldwide, features modest genetic progress due to long selection cycles and the extent of the non-additive genetic variance associated with its autotetraploid genome. METHODS: To improve the efficiency of genomic selection in alfalfa, we explored the effects of genome parametrization (as tetraploid and diploid dosages, plus allele ratios) and SNP marker subsetting (all available SNPs, only genic regions, and only non-genic regions) on genomic regressions, together with various levels of filtering on reading depth and missing rates. We used genotyping by sequencing-generated data and focused on traits of different genetic complexity, i.e., dry biomass yield in moisture-favorable (FE) and drought stress (SE) environments, leaf size, and the onset of flowering, which were assessed in 143 genotyped plants from a genetically broad European reference population and their phenotyped half-sib progenies. RESULTS: On average, the allele ratio improved the predictive ability compared with other genome parametrizations (+7.9% vs. tetraploid dosage, +12.6% vs. diploid dosage), while using all the SNPs offered an advantage compared with any specific SNP subsetting (+3.7% vs. genic regions, +7.6% vs. non-genic regions). However, when focusing on specific traits, different combinations of genome parametrization and subsetting achieved better performances. We also released Legpipe2, an SNP calling pipeline tailored for reduced representation (GBS, RAD) in medium-sized genotyping experiments.

Genomic prediction and allele mining of agronomic and morphological traits in pea (Pisum sativum) germplasm collections.

Well-performing genomic prediction (GP) models for polygenic traits and molecular marker sets for oligogenic traits could be useful for identifying promising genetic resources in germplasm collections, setting core collections, and establishing molecular variety distinction. This study aimed at (i) defining GP models and key marker sets for predicting 15 agronomic or morphological traits in germplasm collections, (ii) verifying the GP model usefulness also for selection in breeding programs, (iii) investigating the consistency between molecular and phenotypic diversity patterns, and (iv) identifying genomic regions associated with to the target traits. The study was based on phenotyping data and over 41,000 genotyping-by-sequencing-generated SNP markers of 220 landraces or old cultivars belonging to a world germplasm collection and 11 modern cultivars. Non-metric multi-dimensional scaling (NMDS) and an analysis of population genetic structure indicated a high level of genetic differentiation of material from Western Asia, a major West-East diversity gradient, and quite limited genetic diversity of the improved germplasm. Mantel's test revealed a low correlation (r = 0.12) between phenotypic and molecular diversity, which increased (r = 0.45) when considering only the molecular diversity relative to significant SNPs from genome-wide association analyses. These analyses identified, inter alia, several areas of chromosome 6 involved in a largely pleiotropic control of vegetative or reproductive organ pigmentation. We found various significant SNPs for grain and straw yield under severe drought and onset of flowering, and one SNP on chromosome 5 for grain protein content. GP models displayed moderately high predictive ability (0.43 to 0.61) for protein content, grain and straw yield, and onset of flowering, and high predictive ability (0.76) for individual seed weight, based on intra-population, intra-environment cross-validations. The inter-population, inter-environment assessment of the models trained on the germplasm collection for breeding material of three recombinant inbred line (RIL) populations, which was challenged by much narrower diversity of the material, over eight-fold less available markers and quite different test environments, led to an overall loss of predictive ability of about 40% for seed weight, 50% for protein content and straw yield, and 60% for onset of flowering, and no prediction for grain yield. Within-RIL population predictive ability differed among populations.