Example

Goal: impute a full genotype from just HLA-A, HLA-B and HLA-DRB1.

Output loci: A, B, DRB1 (or tick more to have them imputed)

Population: CAU

Typing (GL‑string):

A*02:01+A*02:01^B*40:01+B*57:01^DRB1*04:01+DRB1*07:01

Here '+' joins the two alleles at each locus and '^' separates the loci. The person is homozygous at HLA-A (A*02:01 on both chromosomes) and heterozygous at HLA-B and HLA-DRB1.

What you get back: ranked complete genotypes (Genotypes tab), the haplotype pairs that compose them (Haplotype couples tab), and per-population haplotype probabilities (Haplotype separate tab). Because only three loci were typed, the remaining loci are inferred from Caucasian haplotype frequencies, so probability is spread across several plausible genotypes.

Goal: phase a classical 10/10 (five-locus) typing and obtain haplotype probabilities.

Output loci: A, B, C, DRB1, DQB1

Population: CAU (or a detailed code such as NAMER)

Typing (GL‑string):

A*01:01+A*03:01^C*07:01+C*07:02^B*08:01+B*07:02^DRB1*03:01+DRB1*15:01^DQB1*02:01+DQB1*06:02

All five classical loci are typed at high resolution. With this much information the imputation is sharply constrained: a small number of haplotype pairs dominate, and the top genotype usually carries most of the probability. This case includes two well-known European haplotypes, so the phasing is strongly supported.

Goal: work with the extended nine-locus panel including DQA1, DPA1, DPB1 and DRB3/4/5.

Output loci: A, B, C, DRB1, DQA1, DQB1, DRB3/4/5, DPA1, DPB1

Population: CAU

Typing (GL‑string):

A*02:01+A*24:02^C*03:03+C*07:01^B*15:01+B*18:01^DRB1*04:01+DRB1*11:01^DQA1*03:01+DQA1*05:05^DQB1*03:01+DQB1*03:02^DPA1*01:03+DPA1*01:03^DPB1*02:01+DPB1*04:01

Note that DRB3/4/5 is supplied as one locus carrying whichever of DRB3, DRB4 or DRB5 is present on each chromosome (here a DRB3 and a DRB4). If a chromosome carries none of these genes, that allele is simply left out and treated as a null. DPA1 here is homozygous (DPA1*01:03 on both chromosomes).

Goal: only HLA-A, HLA-B and HLA-DRB1 were typed, but you want a full nine-locus prediction.

Output loci: tick all nine

Population: AFA (or a detailed code such as AAFA)

Typing (GL‑string):

A*30:01+A*68:02^B*42:01+B*53:01^DRB1*03:02+DRB1*11:01

You type three loci and request nine. ML-GRIM fills in C, DQA1, DQB1, DRB3/4/5, DPA1 and DPB1 from the African American haplotype frequencies. Expect the probability to be distributed over more genotypes than in the fully typed cases — the untyped loci genuinely have several likely values. To sharpen the result, type additional informative loci (HLA-C and HLA-DRB1 partners help most).

Goal: handle typing exactly as it often arrives from a lab — some loci at high resolution, some at low resolution, some ambiguous.

Output loci: A, B, C, DRB1, DQB1

Population: API (or a detailed code such as JAPI)

Typing (GL‑string):

A*24:02+A*02:01/A*02:07^C*01:02+C*03:04^B*54:01+B*40:01^DRB1*04:05+DRB1*09:01^DQB1*04:01+DQB1*03:03

The second HLA-A allele is ambiguous: A*02:01/A*02:07 means it is either A*02:01 or A*02:07 (the / operator marks allelic ambiguity). GRIMMARD considers both possibilities and uses the population frequencies to weigh them. You can mix in lower-resolution forms the same way — antigen-level codes (e.g. A2), NMDP multiple-allele codes (e.g. A*02:AB), or G/P groups (e.g. A*02:01P) — and py-ard will normalize them before imputation.

Goal: the individual is of uncertain or mixed ancestry, so you want to see how a typing is supported under more than one population.

Output loci: A, B, C, DRB1, DQB1, DPB1

Population: select several, e.g. CAU and HIS and AFA

Typing (GL‑string):

A*01:01+A*02:01^C*06:02+C*07:01^B*57:01+B*08:01^DRB1*07:01+DRB1*03:01^DQB1*03:03+DQB1*02:01

With multiple populations selected, the Haplotype separate tab reports each haplotype's probability under every chosen population, so you can see which population best explains the typing and how the ranking shifts between them.