Alleles not equal on tsinfer verify, v 0.2.0

[stsmall]

Hi,
I am working with the tsinfer version 0.2.0. When running verify, I get an error about alleles not being equal. I checked that all my sites have only 2 alleles and that the ancestral allele matches either the REF or ALT. I also removed all singletons, not sure if this was advised or not. I used this script (https://github.com/stsmall/Kiribina_Folonzo/blob/debug/vcf2tsinfer.py) which was a modification of the one in the tutorial. tsinfer ran to completion without error.
thanks for the help!
@stsmall

$ tsinfer list chrX.trees
path = chrX.trees
size = 519.0 MiB
edges = 9215138
trees = 910682
sites = 2160904
mutations = 2124826

$ tsinfer verify chrX.samples chrX.trees
2021-04-22 10:36:02,999 [47534] CRITICAL root: Traceback (most recent call last):
File "tsinfer", line 8, in
sys.exit(main())
File "tsinfer/main.py", line 5, in main
cli.tsinfer_main()
File "tsinfer/cli.py", line 499, in tsinfer_main
args.runner(args)
File "tsinfer/cli.py", line 238, in run_verify
tsinfer.verify(samples, ts, progress_monitor=args.progress)
File "tsinfer/inference.py", line 181, in verify
raise ValueError(f"alleles not equal: {var1.alleles} != {var2.alleles}")
ValueError: alleles not equal: ('C', 'T') != ('C',)

[hyanwong]

Ah, it looks like these are monomorphic sites, right? Perhaps we forgot to test how those are handled in verify(). I suspect you can ignore this until we fix it, though.

[stsmall]

ugh, yep. Monomorphic sites. I am sorry that I missed that.

[hyanwong]

No problem, we should handle this anyway.

[hyanwong]

Here's a quick test case that current fails:

ts = msprime.sim_ancestry(3, ploidy=1, sequence_length=10, random_seed=123)
ts = msprime.sim_mutations(ts, rate=0.5, model="binary", random_seed=1)
sample_data = tsinfer.SampleData.from_tree_sequence(ts, use_sites_time=False)
is_monomorphic = np.all(np.diff(sample_data.sites_genotypes[:], axis=1)==0, axis=1)
assert len(is_monomorphic) == sample_data.num_sites
assert np.any(is_monomorphic)
ts_inf = tsinfer.infer(sample_data)
tsinfer.verify(sample_data, ts_inf)

Giving:

ValueError: alleles not equal: ('0', '1') != ('0',)

[hyanwong]

The new msprime mutation model is also not guaranteed to give the non-ancestral alleles in the same order as input. I'm working on a fix.

Edit - here's an example:

ts = msprime.sim_ancestry(3, ploidy=1, sequence_length=10, random_seed=123)
ts = msprime.sim_mutations(ts, rate=0.2, random_seed=1)
sd = tsinfer.SampleData.from_tree_sequence(ts, use_sites_time=False)
ts_inf = tsinfer.infer(sd)
has_alt_order = False
for v1, v2 in zip(sd.variants(), ts_inf.variants()):
    if set(v1.alleles) == set(v2.alleles) and v1.alleles != v2.alleles:
        has_alt_order = True
assert has_alt_order
tsinfer.verify(sd, ts_inf)

Currently giving:

ValueError: alleles not equal: ('C', 'T', 'A') != ('C', 'A', 'T')

[hyanwong]

@stsmall - if you have a moment could you check that the latest GitHub version of tsinfer now works with your files with the monomorphic sites?