Thread by @jakphd

Huge congrats to @jdbackman, Manuel Ferreira, @gabecasis, @RegeneronDNA, and co. on @Nature 450k exome @uk_biobank paper 🎉🎉🎉

Especially exciting to see ALL ancestries analyzed (not just EUR like most other UKB exome papers thus far)
nature.com/articles/s4158…

Across 454k individuals, they found 12.3M variants (99.6% MAF<1%). This is:
- 1.3x the coding variation in TOPMed & gnomAD combined
- 8x what could be found from TOPMed imputation of UKB (info >0.3)

Testing both individual variants and burden tests across 3702 binary and 292 quantitative traits (2.3B tests) using REGENIE, they found 8865 associations (Bonerroni P<2.18e-11) across 564 genes, 492 traits, and 2283 gene-trait pairs.

Often forgotten in RV analyses is whether a RV signal is explained by LD with a common variant.

To test for this, they performed a GWAS using imputed data and conditioned the associations on top common variants of which 91% were still significant after conditioning.

Replication is always important but can be difficult with rare variants, partially due to lacking additional large cohorts not used in the initial discovery. With help from a smaller cohort of 133k exomes from @GeisingerHealth, 81% of associations replicated.

Although many different allele frequency thresholds were used for burden tests, they did find 69 (15 novel) associations discovered solely through burden tests of singleton variants (MAF ~1e-6).

Looking for protective associations differs significantly between binary and quantitative traits.
- 131 protective associations in quantitative traits
- 5 in binary traits

Unsurprisingly, the majority of traits with a RV signal have more common variants signals. However, 20 traits only had rare variant associations, of which the signals all came from CHIP genes. These associations were indeed correlated with age and had low AB (>35%; i.e. somatic)

To help pinpoint likely genes in GWAS loci, they ran GWAS for each of the 492 traits with a RV association finding >107k independent loci. After conditioning on GWAS loci, RV associations were:
- 59x enriched for the nearest gene
- 11.4x for genes within 1MB of GWAS loci

So despite what some call an arbitrary decision to assign the nearest gene to a GWAS locus, this really isn't a terrible decision as corresponding RV associations are 59.4x enriched in the nearest gene.

For those attending #ASHG21, the first author, @jdbackman, is giving a #plenary covering the results of this manuscript during plenary session 1 at 1:20pm -1:40pm EST.

Perhaps most importantly, the full associations will be publicly available via @GWASCatalog (they have already been deposited and are currently undergoing quality checks).