1 An important challenge in genetics is to be able to predict complex traits accurately. Despite 2 recent advances, prediction accuracy for most complex traits remains low. Here, we used the 3 Drosophila Genetic Reference Panel (DGRP), a collection of 200 lines with whole-genome 4 sequences and deep RNA sequencing data, to evaluate the usefulness of using high-quality gene 5 expression levels compared to relying on genotypes for predicting three complex traits. We 6 found that expression levels provided higher accuracy than genotypes for starvation resistance, 7 similar accuracy for chill coma recovery, and lower accuracy for startle response. Models 8 including both genotype and expressions levels did not outperform the best single component 9 model. However, accuracy increased considerably for all the three traits when we included 10 another layer of information, i.e., gene ontology (GO). We found that a limited number of GO 11 terms, some of which had a clear biological interpretation, were strongly predictive of the traits. 12In summary, this study shows that integrating different sources of information can improve 13 prediction accuracy, especially when large samples are not available. 14 2006;Hayes et al. 2009). 32 33 GS has been applied to many agricultural species, the first being livestock. Despite the fact that 34 GS has revolutionized animal breeding, the results have varied greatly between species. In dairy 35 cattle, which has the largest and highest quality reference populations, the accuracy of genomic 36 estimated breeding values (gEBVs) has been ~0.7-0.8 for many traits. This high accuracy 37combined with the reduced generation interval has doubled the rate of genetic gain since the 38 implementation of GS in breeding programs (Meuwissen et al. 2016). However, in beef cattle 39 and pigs, for example, the reference populations are much smaller within each breed, resulting in 40 lower accuracy. To overcome this issue, multi-breed reference populations have been used. 41