Know more

The main objective of genomic prediction is to use genomic variation, generally single nucleotide polymorphisms (SNP) to predict phenotypes. However, certain complex traits are difficult to predict. In parallel, functionnal genomic knowledge obtained from omics studies are promising, allowing a better understanding of the underlying cell mechanisms and the etiology of traits.

The integration of this information in genomic prediction models could potentially lead to a better understanding of complex traits. Bayesian models, through their use of priority, seem to be a means for directly introducing known functional information into genomic prediction models for complex traits, to potentially improve  prediction quality and to better understand genomic architecture. In particular, BayesRC (MacLeod et al., 2016) categorizes SNP in multiple disjointed annotations, so that the QTL propositions in each can vary. Although this model gives good results, it is limited by the non-overlapping nature of the annotations, which prevent the SNP from belonging to more than one annotation. When the number of available annotations increases, this becomes an important limitation, preventing us from making the most of the information available.

We have designed two new Bayesian models that include these multi-annotated SNP, by attributing a cumulative (BayesRC+) or a preferential (BayesRC..) contribution through annotation categories. These two models are implemented in BayesRCO, an open source Fortran 90 software (https://github.com/fmollandin/BayesRCO). We will apply these models on data produced by the GENE-SWitCH project.

Contact :

• Fanny Mollandin