Search

Additional file 5: Figure S5. Stability of the relationships between change in predictorrank and muti-omic prediction advantage for traits measured at Orleans. The relationship is measured as the correlation coefficient between change in rank and relative advantage in prediction with the multi-omic model over the single-omic counterpart. The stability of the relationship is evaluated with respect to the categorization of eQTLs into cis, trans, or not eQTLs, being defined according to windows of increasing size in bp around the SNPs detected with the multi-locus model.

Additional file 6: Figure S6. Relationship between change in predictor rankand muti-omic prediction advantage for traits measured at Savigliano. Regression across traitsmeasured at Savigliano between average change in predictor rank and advantage in performance of the multi-omic model (G+T) over the single-omic counterpart. Ranks were computed from squared effects of features in the ridge-regression models. The top panel (A) shows the regression obtained with the eQTLs (trans-eQTLs on the left, cis-eQTLs in the middle, and SNPs not detected as eQTL on the right). The bottom panel (B) shows the regression obtained with the regulated transcripts (trans on the left, cis in the middle, and not found to be associated with eQTLs on the right).

Additional file 1: Figure S1. Heritability distribution across wavelengths for each pre-process. Figure S2. Co-inertia analysis between SNP and multi-year wood and leaves NIRS matrices. Figure S3. Co-inertia analysis between SNP, single and multi-year wood and leaves NIRS matrices. Figure S4. Distribution of phenomic prediction predictive ability per pre-process and per trait. Figure S5. Predictive ability of phenomic prediction with a single vs both tissues. Figure S6. Comparison of methods for genomic prediction. Figure S7. Distribution of phenomic prediction predictive ability over 15 traits in each half-diallel cross. Figure S8. Predictive ability of combined vs genomic prediction models.

Additional file 4: Figure S4. Change in rank of the eQTLs and their corresponding targeted transcripts. Scatter plots of the changes in rank of the eQTLs and their corresponding targeted transcripts for each trait. Ranks were computed from squared effects of features in the ridge-regression models. The red and blue dots represent the regulation in transand cis, respectively.

Additional file 1: FigureS1. Proportion and score of cis- and trans-eQTLs. Proportions(A) and scores (B) of cis- and trans-eQTLs detectedwith the multi-locus approach. Scores correspond to the -log10 ofthe p-value of the test.

Additional file 3: Figure S3. Variation of the change in rank of the eQTLs and targeted transcripts. Barplot of the variance of change in rank of eQTLs (top) and targeted transcripts (bottom) for each trait (panel) and site (dark grey: Orleans ; light grey: Savigliano). Ranks were computed from squared effects of features in the ridge-regression models.

Additional file 2: Figure S2. Comparison between the rank of predictors across single- and multi-omic models. Panels A and C represent the boxplots of SNPs and transcripts rank, according to the traits and sites. Features are grouped into the following categories determined after the eQTL analysis: cis or trans eQTL or regulated transcripts and not eQTLs. Panels B (SNPs) and C (transcripts) represent the correlations between the ranks of predictors across single- and multi-omic models, splitting the predictors into the same previous categories, determined from the eQTL analysis. Ranks were computed from squared effects of features in the ridge-regression models.

International audience; Genotypic variations in fruit tree architectural and functional traits have been reported but their genetic bases and the existence of common geneticdeterminants have rarely been investigated in wide genetic diversity. We made use of innovative in-eld high-throughput phenotyping on anassociation panel of 241 genotypes to quantify architectural and functional traits on 4-years old apple trees. T-Lidar technology was used for estimatingarchitectural traits (number of branches, tree height, STAR…), airborne thermal and multispectral imagery for assessing canopy temperature (a proxyof transpiration rate) and vegetation indices (NDVI, MCARI2, GNDVI). This dataset was complemented with yield-related traits (number of fruits, averagefruit weight). We explored genome wide associations on 16 traits with high density (275K) genotypic data. GWAS was run with a multi-locus mixedmodel using genetic kinship matrix and assuming different thresholds for SNP signicance: (i) the Bonferroni correction (Bonfth, -logpval > 6.7), (ii) acorrection based on an estimated number of independent SNPs (Bonfcor, -logpval > 5.7) and (iii) value based on common practices (-logpval = 5).Signicant SNPs with -logpval > 5 were detected for most traits with a number of them displaying -logpval above Bonfcor and above Bonfth. One to foursignicant SNPs were identied by variable with a proportion of variance explained by each SNP ranging from 6% to 14% and the total proportion ofvariance explained by all signicant SNPs exceeding up to 30% for the number of branches, STAR, MCARI2, number of fruit and crop load. SNPscontrolling different traits were detected on LG9 (height, STAR, volume) and LG13 (volume, MCARI2, NDVI, GNDVI). Analyses of the LD between SNPsallowed determining possible colocations within regions. Notably, some signicant SNPs were localized in genomic regions previously reported as QTLsfor similar traits in bi-parental population. This suggests their robustness and encourages further investigations.