Genetic correlation results for heritable phenotypes in the UK Biobank
We are thrilled to announce the release of downloadable genetic correlation results for the significantly heritable phenotypes in our UK Biobank application (see ‘Defining a set of significant SNP-heritability results’ in our recent heritability results blog post). We have genetic correlation results for both sexes, as well as male/female-specific subsets. The results are available now to explore and download at ukbb-rg.hail.is.
For this genetic correlation release, LD-score correlation (using the LD-score regression software: github.com/bulik/ldsc) was applied to all of the phenotypes available in our application. This encompassed all the phenotypes for which we released summary statistics for our ‘Round 2 GWAS’. Importantly, the Round 2 GWAS results uses a larger sample size (n=361,194) than Round 1 (n=337,199) by using a less strict inclusion criteria based on principal components. We also dramatically increased the number of phenotypes in Round 2 from 2,419 to 4,236 - see here for details.
In addition to the raw results, we have also created a set of visualisation and dimensionality reductions which you can take a look at in the plots section of our website menu. We first display the large genetic correlation matrix both alphabetically and sorted by similarity using hierarchical clustering - hover and zoom in on blocks to check out traits that group together!
Further down the menu, you can examine linear and non-linear dimensionality reduction using principal components analysis, with both diffusion map and UMAP values projected into two dimensions.
For details describing how these methods work, see en.wikipedia.org/wiki/Diffusion_map and arxiv.org/abs/1802.03426, and the public repository for our implementation and plotting regime.
Results availability
We have created a browser for users to peruse results: ukbb-rg.hail.is/rg_browser. You can explore collections of phenotypes with the search or dropdown menus and use the sliders to restrict to rg and p-value ranges. The entirety of the output from the LD-score correlation regressions can be included using the column visibility button and downloaded with your selections.
In the website, we restrict our attention to the most heritable phenotypes. We define these ‘significant’ phenotypes as those that have medium or high confidence based on the latest heritability analysis and for which the associated Z-score for testing non-zero heritability is greater than 4. Please see our heritability blog post and results website for details.
While navigating to the browser, you can select a phenotype of interest to take you to a phenotype specific page showing an interactive volcano plot and phenotype specific table. For example, let’s take a look at preference for posh coffee.
Hovering over points you can examine the phenotypes pairs with their genetic correlations and associated p-values. You can click on the key to remove classes of traits, and double click to isolate just that class. Finally, by clicking the tabs you can easily jump between the both sexes and sex specific analysis results for the trait you’re looking at.
The full collection of genetic correlations without cutoffs or restrictions guided by the heritability analysis are also available for download.
We have made some small alterations to flags in the ldsc code to enable incorporation of many more phenotypes on input and output in https://github.com/astheeggeggs/ldsc, which in combination with the LDSC sumstats files, can be used to regenerate these results or modified to run further analyses for your application.
All of the code used to generate the results and our browser are available on github, including readme files with step-by-step instructions to create your own version of the website!
In addition to providing tidbits of the latest round of heritability results, the GWAS bot @sbotgwa will start spouting information about genetic correlations in addition to the usual manhattan plot fare.
If you have any questions, comments, or spot any bugs or typos, do feel free to contact us at nealelab.ukb@gmail.com and we’ll get back to you as soon as possible. Thanks for everyone’s continued interest in our work and blog posts, we really hope our site is useful and you enjoy having a browse!
Authored by Duncan Palmer