1st Plenary #ASHG20 session: Meredith Course identified a human-specific 69bp repeat expansion in the last exon of WDR7.
This repeat was associated with ALS (repeats in ALS cases are on average longer in ALS cases than controls). Longest repeat observed was 86bp. #ASHG20
Observed periodicity in the WDR7 repeat and found that it expands in multiples of 2 in the 3' - 5' direction. #ASHG20
Initial analysis was done in a EUR cohort, so repeated the analysis in non-EUR cohorts, finding population specific repeats in Han Chinese descent and AFR ancestry #ASHG20
Found that the WDR7 repeat first appears in the great apes. Looking at ancient DNA, found the repeat was present in both Neanderthal and Denisovans.
The function of the repeat can form microRNAs #ASHG20
In case you missed it, I presented results on behalf of @Regeneron from a trans-ancestry #COVID19 meta-analysis of common and rare variants + gene burden tests in >883k imputed samples and >592k exomes. #ASHG20
Using REGENIE developed by @joellembatchou and @marchini (SAIGE gave us some bizarre results with rare variants) to run our common and rare variant GWASes, we found 2 loci associated with susceptibility and 3 loci with hospitalization. #ASHG20
Curiously, despite losing ~75% of the cases, we found more loci with hospitalization than using all COVID19 positive individuals - @covid19_hgi sees the same pattern.
One might suspect a severe COVID19 GWAS would be even more powerful at the same sample size #ASHG20
Up next in the first #ASHG20 plenary session is my former Daly lab colleague, @HHeyne, presenting work from @FinnGen_FI on "Recessive effects of 82,516 coding variants in 176,899 Finns."
Bottleneck in Finland makes the Finnish population and @FinnGen_FI ideal for understanding ultra-rare variants that stochastically rose in frequency #ASHG20
Using SAIGE, ran GWAS using 1) a recessive model and 2) additive model on all coding variants in @FinnGen_FI.
For the majority of coding variants, additive model performs better. However, the recessive model performed 2x better for some variants. #ASHG20
Up next is @joellembatchou from @Regeneron presenting REGENIE - a computationally efficient whole genome regression for quantitative and binary traits #ASHG20
Both BOLT and SAIGE are fantastic LMMs for quantitative and binary traits respectively, but have high memory requirements and long computational times. #ASHG20
Noticed inflation in Betas as MAF decreases in SAIGE and becomes worse with larger case/control imbalance - meaning for rare variants, SAIGE produces nonsensical results. REGENIE resolves this issue #ASHG20