In recent years, the cost of sequencing and genotyping genomes has fallen dramatically.  This has led to a recent deluge of genetic datasets, where more than a million people are expected to be genotyped this year alone. An important benefit from large number of genotypes and measured traits is that it may allow us to train predictors that can predict disease risk from individual genotype.  Most genetic tests that are currently used in the clinic rely on rare genetic variants with large effects on disease risk, therefore only predicting a small amount of disease cases.  In contrast, given enough training data predictors trained on the entire genome can predict the entire heritable component yielding more accurate predictions for of many diseases and heritable traits. These include heritable diseases like type-2 diabetes, coronary artery disease, rheumatoid arthritis, multiple sclerosis, breast cancer, and many more where preventative measures and early treatment can have a significant impact on prognosis.

Polygenic risk scores are a promising method that can train on large genetic datasets for predicting disease risk (Dudbridge, PloS Genet 2013; Chatterjee et al., Nat Genet 2013). As a more accurate predictor, we propose a new Bayesian method, LDpred, which both in theory and in simulations outperforms the standard polygenic risk score approach, particularly at large sample sizes. Prediction accuracy, as measured by squared correlation, increased from 20.1% to 25.3% in a large schizophrenia data set and from 9.8% to 12.0% in a large multiple sclerosis data set. The advantage of LDpred over existing methods will grow as sample sizes increase.

About the speaker:

Bjarni Vilhjalmsson obtained his B.Sc. in Computer Science/Mathematics in 2003 and M.Sc. in Bioinformatics in 2006 from the University of Copenhagen.  He obtained his Ph.D. in Computational Biology from the University of Southern California in 2011, after which he went on to become a postdoctoral fellow with Prof. Alkes Price at Harvard University.  In the fall of 2014 Bjarni moved back to Denmark and is currently supported by a DFF individual postdoctoral grant at the Bioinformatics Research Centre at Aarhus University.   Bjarni has several ongoing international collaborations, including collaboration with deCODE genetics, and former colleagues at Harvard and the Broad Institute.