Output

kmate run writes a tab-separated table with a header, one row per variant in the panel:

chrom  pos  ref_len  alt_len  alt_freq  info  n_called  se

column	meaning
chrom, pos	position, 1-based
ref_len, alt_len	length of the REF and ALT alleles
alt_freq	frequency of the ALT allele in the pool, 0-1
info	how much haplotype mass had a genotype call here
n_called	how many founder haplotypes had a genotype call here
se	binomial standard error of alt_freq from n_called

Alongside it, *.h_per_chrom.npz holds the estimated founder-haplotype mixture: one value per haplotype, summing to 1.

---

There is no variant-type column

Derive it from the lengths:

type	rule
SNP	ref_len == 1 and alt_len == 1
indel	ref_len != alt_len and max(ref_len, alt_len) < 50
structural variant	max(ref_len, alt_len) >= 50

import pandas as pd
df = pd.read_csv("MYPOOL_Chr1.tsv", sep="\t")
snp = (df.ref_len == 1) & (df.alt_len == 1)
sv  = df["ref_len", "alt_len"](/Tatianabellagio/kMate/wiki/"ref_len",-"alt_len").max(axis=1) >= 50

---

Comparing samples

Every sample run against the same panel produces the same rows in the same order, so they stack directly:

import pandas as pd
a = pd.read_csv("POOL_A_Chr1.tsv", sep="\t")
b = pd.read_csv("POOL_B_Chr1.tsv", sep="\t")
delta = b.alt_freq - a.alt_freq          # row-wise, no merge needed

Do not merge on chrom and pos. A position can carry several variants (a SNP and an indel starting at the same base), so a position key matches the wrong allele. Join row-wise, or on all four of chrom, pos, ref_len, alt_len.

---

Judging an estimate

• n_called is the support behind each estimate: a frequency from few haplotypes is weak regardless of se. Filter on it.
• alt_freq is a frequency among called haplotypes. Founder haplotypes missing at a variant are excluded rather than counted as reference.
• Check the founder-haplotype mixture before trusting per-variant numbers. Collapse onto a handful of haplotypes where many were expected indicates a problem upstream: coverage, the panel, or --unit.