We have developed Genomic Feature Linear Mixed Models for predicting quantitative trait phenotypes from high resolution genomic polymorphism data. Genomic features are regions on the genome that are hypothesized to be enriched for causal variants affecting the trait. Several genomic feature classes can be formed based on previous studies and different sources of information including genes, chromosomes, biological pathways, gene ontologies, sequence annotation, prior QTL regions, or other types of external evidence. Using prior information on genomic features is important because prediction is difficult for populations of unrelated individuals when the number of causal variants is low relative to the total number of polymorphisms, and causal variants individually have small effects on the traits. The models were implemented using likelihood or Bayesian methods.
We have developed Genomic Feature Best Linear Unbiased Prediction (GFBLUP) models. We have extended these models to include multiple features and multiple traits. Different genetic models (e.g. additive, dominance, gene by gene and gene by environment interactions) can be specified.
We have developed Bayesian multiple Genomic Feature and Trait models. The models are implemented using an empirical Bayesian method that handles multiple features and multiple traits. The models were implemented using spectral decomposition that plays an important computational role in the Markov chain Monte Carlo strategy. This is a very flexible and formal statistical framework for using prior information to decompose genomic (co)variances and predict trait phenotypes.
The premise of the Genomic Feature models presented above is that genomic features are enriched for causal variants affecting the traits. However, in reality, the number, location and effect sizes of the true causal variants in the genomic feature are unknown. Therefore we have developed and evaluated a number of SNP set tests derived from a standard Genomic BLUP model. These approaches are computationally very fast allowing us to rapidly analyze different layers of genomic feature classes to discover genomic features potentially enriched for causal variants. Results from these analyses can be built into the above mentioned prediction models.
| Package: | qgg |
| Type: | Package |
| Version: | 1.0 |
| Date: | 2015-10-21 |
| License: | GPL-3 |
Mapping Variants to Gene Ontology Categories Improves Genomic Prediction for Quantitative Traits in Drosophila melanogaster. Under review Genetics (2016). Edwards SM, S<U+00F8>rensen IF, Sarup P, Mackay TF, S<U+00F8>rensen P.
Genomic BLUP Derived Set Tests Identify Genetic Variants Associated with Schizophrenia in Functionally Associated Biological Processes. Under review, Genetics (2015). Rohde PD, Demontis D, The GEMS Group, B<U+00F8>rglum AD, S<U+00F8>rensen P.
Partitioning of Genomic Variance Reveals Biological Pathways Associated with Udder Health and Milk Production Traits in Dairy Cattle. GSE (2015) 47:60. Edwards SM, Thomsen B, Madsen P, S<U+00F8>rensen P.
Increased Prediction Accuracy using a Genomic Feature Model Including Prior Information on Quantitative Trait Locus Regions in Purebred Danish Duroc Pigs. BMC Genetics (2016) 17:11. Sarup P, Jensen J, OstersenT, Henryon M, S<U+00F8>rensen P.