Using Whole-Genome Sequence Data to Predict Quantitative Trait Phenotypes in Drosophila melanogaster

Abstract

Predicting organismal phenotypes from genotype data is important for plant and animal breeding, medicine, and evolutionary biology. Genomic-based phenotype prediction has been applied for single-nucleotide polymorphism (SNP) genotyping platforms, but not using complete genome sequences. Here, we report genomic prediction for starvation stress resistance and startle response in Drosophila melanogaster, using ∼2.5 million SNPs determined by sequencing the Drosophila Genetic Reference Panel population of inbred lines. We constructed a genomic relationship matrix from the SNP data and used it in a genomic best linear unbiased prediction (GBLUP) model. We assessed predictive ability as the correlation between predicted genetic values and observed phenotypes by cross-validation, and found a predictive ability of 0.239±0.008 (0.230±0.012) for starvation resistance (startle response). The predictive ability of BayesB, a Bayesian method with internal SNP selection, was not greater than GBLUP. Selection of the 5% SNPs with either the highest absolute effect or variance explained did not improve predictive ability. Predictive ability decreased only when fewer than 150,000 SNPs were used to construct the genomic relationship matrix. We hypothesize that predictive power in this population stems from the SNP–based modeling of the subtle relationship structure caused by long-range linkage disequilibrium and not from population structure or SNPs in linkage disequilibrium with causal variants. We discuss the implications of these results for genomic prediction in other organisms.

Toward genomic prediction from whole-genome sequence data: impact of sequencing design on genotype imputation and accuracy of predictions

“…The results differ both from those reported by Meuwissen and Goddard (2010), based on simulated data, and those from Ober et al (2012), based on real data from Drosophila. Meuwissen and Goddard (2010) simulated QTL allele frequencies following that expected under a neutral model.…”

“…This increase in accuracy has not been observed in real data as yet. Ober et al (2012) found no increase in accuracy of predictions for quantitative traits in 157 inbred lines of Drosophila melanogaster, when comparing predictions from a dense SNP panel or the full-genome sequence. However, the small size of that data set makes it difficult to draw definitive conclusions about the value of full-genome sequence data in genomic predictions-as the effect of the causative mutations on the quantitative traits may be very small, given the genetic architecture observed for many quantitative traits (for example, Kemper et al (2011) and Stahl et al (2012)), large numbers of individuals will still be required to estimate these effects accurately.…”

“…et al (2012), on the other hand, observed no advantage (over dense SNP data) of using whole-genome sequence data for genomic predictions of starvation response in D. melanogaster. In fact, it is difficult to compare our results to those of Ober et al (2012), because the valid comparison depends on an assumption regarding the frequency of QTL alleles for starvation response. If we assume that starvation response in Drosophila has been under strong selection, and QTL alleles are at MAF o1%, then our results are not consistent with the results of Ober et al (2012), as we did see an advantage of the sequence data in this situation.…”

Effects of larval crowding on quantitative variation for development time and viability in Drosophila melanogaster

“…Variance components (σ 2 ) were calculated from the full models fitting genotype, genotype‐by‐sex and replicate vial as random effects. Broad‐sense heritabilities can be calculated as variance due to genotype divided by the full variance (Ober et al., 2012) and are marked in the table with bold numbers. *p < .05, **p < .01, ***p < .001.…”

“…For the log‐transformed DT data, the model included the fix effect of sex and the random effect of replicate vial as well: Y ijkl  =  μ  + L i   + S j   + L i S j   + V k   +  ε ijkl . Broad‐sense heritability (repeatability) was calculated as H 2  =  ${\mathrm{normal\sigma }}_{\mathrm{normalL}}^{2}false/false({\mathrm{normal\sigma }}_{\mathrm{normalL}}^2+{\mathrm{normal\sigma }}_{\mathrm{normalLS}}^2+{\mathrm{normal\sigma }}_{\mathrm{normalV}}^2+{\mathrm{normal\sigma }}_{\mathrm{normalresidual}}^2$) for DT (Ober et al., 2012), and as 4*${\mathrm{normal\sigma }}_{\mathrm{normalL}}^{2}false/false({\mathrm{normal\sigma }}_{\mathrm{normalL}}^2$   +  π 2 /3) for EAV (Davies, Scarpino, Pongwarin, Scott, & Matz, 2015), where ${\mathrm{\sigma }}_{\mathrm{L}}^2$ is the variance of the random line effect and π 2 /3 is the variance of the logistic distribution. To eliminate the scale effects and therefore have a comparable measure of variability of the traits across the densities, we used coefficients of variation (CVs; Houle, 1992).…”