,
7 tweets,
3 min read
Read on Twitter
Our latest preprint, led by @beryl_bbc, starts with a mystery: why are there hundreds of people in gnomAD who shouldn't be there? These people carry predicted truncating variants in genes where these are known to cause severe dominant pediatric disease.
Our team manually curated 401 of these variants, and found (as expected) a ton of errors. One of the major error types was what we called a "transcript error", meaning the variant fell on a transcript that looked like it wasn't real, or was very weakly expressed.
So @beryl_bbc devised a systematic approach to annotate variants based on the expression of the affected exon across GTEx tissues, accounting for various biases in short-read data. The basic strategy is summarized in this figure: 
The final metric was called (after some agonizing) the proportion of expressed transcripts or pext score. And it turns out to be broadly useful in separating important bits of genes from unimportant ones - e.g. throwing out lots of common LoFs but few ClinVar variants.
It also improves gene discovery exercises in severe diseases like developmental delay and autism, by making it possible to remove a subset of LoF variants that are weakly expressed and have no disease signal, and flagging highly expressed regions with larger effects.
Interested in using pext? The easiest way to get started is to check out your favorite gene in the fancy new update to the gnomAD browser, which now shows pext across all tissues (thanks to Nick Watts).
You can also download the code from GitHub: github.com/macarthur-lab/… And here's the preprint link again: biorxiv.org/content/10.110… As always, we'd be delighted to hear your comments on the preprint and the results!
