Productionize LD code for gnomAD v4 SNVs/Indels #634
Conversation
|
okay I'm actually okay attaching my name to this now! Ready for review |
| pop_freq = pop_mt.freq[meta_index] | ||
| pop_mt = pop_mt.annotate_rows(pop_freq=pop_freq) | ||
|
|
||
| pop_mt = pop_mt.filter_rows((hl.len(pop_mt.filters) == 0)) |
There was a problem hiding this comment.
Minor comment: this line duplicates line 88
| pop_freq = pop_mt.freq[meta_index] | ||
| pop_mt = pop_mt.annotate_rows(pop_freq=pop_freq) | ||
|
|
||
| pop_mt = pop_mt.filter_rows((hl.len(pop_mt.filters) == 0)) |
There was a problem hiding this comment.
More of a note for future reference:
In my case this line of filter is replaced with a set of filters below,
- Entries: high quality variants (~VQSR)
- Entries: adj filters
- Rows: hl.agg.any(mt.GT.n_alt_alleles() > 0)
| return pop_mt | ||
|
|
||
|
|
||
| def generate_ld_pruned_set( |
There was a problem hiding this comment.
Just a note to confirm, this function is not needed for the purpose of computing LD scores.
| ), | ||
| overwrite, | ||
| ) | ||
| ld = hl.ld_matrix(pop_mt.GT.n_alt_alleles(), pop_mt.locus, radius) |
There was a problem hiding this comment.
In my old script, I ran BlockMatrix.write_from_entry_expr( mt.GT.n_alt_alleles(), tmp_bm_path, mean_impute=True, center=False, normalize=False, overwrite=args.overwrite ), wondering how much difference this will introduce
| ) | ||
| ld = hl.ld_matrix(pop_mt.GT.n_alt_alleles(), pop_mt.locus, radius) | ||
| if data_type != "genomes_snv_sv": | ||
| ld = ld.sparsify_triangle() |
There was a problem hiding this comment.
question: any thoughts in why this shouldn't be applied to all cases?
|
|
||
| l2row = r2_adj.sum(axis=0).T | ||
| l2col = r2_adj.sum(axis=1) | ||
| l2 = l2row + l2col + 1 |
There was a problem hiding this comment.
one more note, I had this line as
r2_diag = checkpoint_tmp(r2_adj.diagonal()).T
l2 = l2row + l2col - r2_diag
Ported code from gnomAD v2. Not running on SVs or for cross-pop analyses