Predicting quantitative traits from genome and phenome with near perfect accuracy

Abstract

In spite of decades of linkage and association studies and its potential impact on human health, reliable prediction of an individual's risk for heritable disease remains difficult. Large numbers of mapped loci do not explain substantial fractions of heritable variation, leaving an open question of whether accurate complex trait predictions can be achieved in practice. Here, we use a genome sequenced population of B7,000 yeast strains of high but varying relatedness, and predict growth traits from family information, effects of segregating genetic variants and growth in other environments with an average coefficient of determination R 2 of 0.91. This accuracy exceeds narrow-sense heritability, approaches limits imposed by measurement repeatability and is higher than achieved with a single assay in the laboratory. Our results prove that very accurate prediction of complex traits is possible, and suggest that additional data from families rather than reference cohorts may be more useful for this purpose.

Powerful decomposition of complex traits in a diploid model using Phased Outbred Lines

“…As a direct consequence of the experimental design, each POL shares one haploid genome with siblings spawned from the same haploid parent. This sharing of half a genome had surprisingly large effects on trait similarity, greatly aiding both trait prediction from relatives 23 and the partitioning of trait variation into its additive, dominant and epistatic components. In contrast, it somewhat restricted our ability to distinguish the weaker effects of individual loci and the calling of those QTLs.…”

“…1a). This sharing of half a genome accounted for surprisingly much of the overall variation in traits 23 , which somewhat restricted our capacity to distinguish contributions from individual alleles and allele pairs from the effect of the genetic background. Nevertheless, our platform provided a cost-efficient framework for calling both additive and nonadditive (dominance and epistasis) QTLs in diploid models.…”

“…QTL calling by linear mixed models, also accounting for population structure, was performed and used as verification. For these, in order to test each QTL, we constructed the realised genetic relationship matrix by discarding the SNPs within the 50kb neighbourhood of the SNP under consideration; these models were fitted with LIMIX 44 as in Märtens, Hallin et al (2015) 23 . Consecutive markers having the same genotype across all individuals were removed for increased computation speed, leaving 10,726 segregating sites 23 .…”

Scan-o-matic: high-resolution microbial phenomics at a massive scale

“…This throughput is orders of magnitude better than what can be achieved by liquid microcultivation. In recent proof-of-principle studies, we have illustrated the importance of simultaneous high-resolution and high throughput analysis, using Scan-o-matic to completely decompose trait variation in diploids into its dominant, epistatic, and additive components , and to predict traits of individuals with near perfect accuracy from their genome and the genome and phenome of relatives (Martens et al 2016). Here, we showed that Scan-o-matic nicely captures the known salt biology of S. cerevisiae, largely uncovering the same cellular functions as earlier found to be important by microcultivation .…”

Deep learning for computational biology

“…As a case in point, the most accurate prediction of gene expression levels is currently made from a broad set of epigenetic features using sparse linear models (Karlic et al , ; Cheng et al , ) or random forests (Li et al , ); how the selected features determine the transcript levels remains an active research topic. Predictions in genomics (Libbrecht & Noble, ; Märtens et al , ), proteomics (Swan et al , ), metabolomics (Kell, ) or sensitivity to compounds (Eduati et al , ) all rely on machine learning approaches as a key ingredient.…”