Tuesday, February 11, 2020
3:30 pm - 4:30 pm
Brian L. Strom Conference Room, Room 701 Blockley Hall, Blockley Hall, 423 Guardian Drive, Philadelphia, PA 19104
Abstract: Genomic prediction for human complex traits using data from genome-wide association studies (GWAS) has attracted great attention, primarily owing to the potential to translate numerous GWAS findings to medical advancements. However, there are some challenges in real data applications, such as the high dimensionality of genetic variants and high polygenicity of complex traits. Here, I present recent work on addressing these challenges. A general statistical guideline is introduced for a wide range of complex traits with varying degrees of polygenicity. In addition, I model GWAS in a random matrix theory framework to theoretically quantify the widespread empirical gap between the prediction accuracy (i.e., incremental R-squared) and signal strength (i.e., heritability). The results of 33 complex traits from different domains in the large-scale UK Biobank database will be illustrated. I will also discuss some ongoing and future work in complex traits prediction, including the cross- ethnic prediction and omics data integration.