Bioinformatics Seminars

Bioinformatics Seminar

Time: 10:45am Tuesdays.
Venue:
Level 7 Seminar Room 2, WEHI1

8 August 2017

Visit the biography of the speaker

Re-evaluation of SNP Heritability in Complex Human Traits

David Balding
Centre for System Genomics, University of Melbourne

"Measuring heritability from SNPs genotyped in unrelated individuals has had a large impact on complex trait genetics since first introduced in 2010. The method has been used to show that common genetic markers carry a large fraction of heritability, distributed across many SNPs genome-wide, thus helping solve the "missing heritability" problem. It has also been used to locate heritability in specific genomic regions, as well as to identify substantial shared heritability across different traits.

By careful modelling and analysis of large genome-wide data sets for many traits we have shown that the mathematical model proposed in the foundational 2010 paper, and widely used since, does not match reality in key aspects. Specifically, the standard model implicitly assigns the same heritability to every SNP a priori. We propose a revised model that uses empirically-supported adjustments based on the properties of the SNP: linkage disequilibrium, population minor allele fraction and genotype quality. Using our revised model we found on average around a 40% increase in SNP-heritability compared with the standard model, thus further reducing the missing heritability gap. However in some cases we found a dramatic reduction. A 2015 paper reported that 80% of heritability was concentrated in DNase Hypersensitivity Sites (DHS), a striking result that attracted much attention because DHS correspond to only about 18% of the genome. In contrast, we estimate that only about 25% of heritability lies in DHS, corresponding to an enrichment factor of about 1.4, slightly less than the 1.6 fold enrichment for genic SNPs.

Our revised model has implications that we will discuss for many other currently-popular methods in complex trait genetics, including methods based on summary statistics rather than individual genotype data."



Search past seminars