broadinstitute / gnomad_chets

Phase of compound heterozygotes in gnomAD

gnomad_chets

Phase of compound heterozygotes in gnomAD

This repository serves as a home for the pipeline used to infer the phase of rare variants in the gnomAD v2 exomes, as reported in our corresponding manuscript (see https://www.nature.com/articles/s41588-023-01608-3), and is coded in Hail 0.2.

The main components of the pipeline can be found in “phasing.py”. Briefly, to infer variant phase, we generate haplotype frequency estimates from genotype counts by applying the expectation-maximization (EM) algorithm (see “get_em_expressions” function) and calculate the probability of two variants being in trans (compound heterozygous, “p_chet”), using the haplotype frequency estimates in a simple equation (“get_em_expr” function).

The remaining scripts in the repository serve to compute the phase of rare variant pairs specifically in the gnomAD and Center for Mendelian Genetics rare disease datasets, and to generate the gnomAD variant co-occurrence look-up tool (see https://gnomad.broadinstitute.org/variant-cooccurrence) and variant co-occurrence counts by gene resource (see https://gnomad.broadinstitute.org/news/2023-03-variant-co-occurrence-counts-by-gene-in-gnomad/). These scripts cannot be run outside of the gnomAD team, as they require access to the individual level data, and are provided for reference only.

gnomad_chets's Issues

Add readme describing how to run scripts

Perform additional analyses to enhance v1 of variant co-occurrence feature

Posted in Slack yesterday by Michael:

1. Benchmarking our phasing accuracy against phasing using programs like Eagle. Perhaps sufficient to just do one chromosome to reduce computational burden.
2. Should we add in gnomAD v2 WGS data to our browser/dataset? Essentially consider only coding regions of the WGS (i.e., treat the WGS as exomes) to help boost sample size.
3. Take a deeper dive into the CMG samples that were incorrectly phased and try to better understand why
4. Test our accuracy against multinucleotide variants as a truth set. This would be particularly interesting since MNVs should be sufficiently close to each other such that recombination doesn’t play a role
5. Better understand how we perform in the scenario of singletons, particularly those that are phased as compound hets.
6. Compile list of genes with compound heterozygous pLOFs. Can we generate new insights into which genes can’t tolerate bi-allelic LOFs?

With additional request by Heidi:
7. Consider including variants deeper into the flanking intron

We'll need to define what we want to have done for v4.