11  Pedigree Reconstruction with repare

Given a set of closely related ancient individuals, it is possible to reconstruct an ancient pedigree. However, while many tools exist to infer pairwise kinship relations in ancient DNA, assembling these relations into a coherent pedigree is a separate and difficult task. repare is a Python tool that takes pairwise kinship relation inferences and reconstructs full pedigrees, while automatically identifying and correcting inconsistent inputs (Huang, Li, and Narasimhan 2026). repare is available at https://github.com/Narasimhan-Lab/repare.

11.1 Background

11.1.1 From kinship to pedigrees

There are numerous methods that can infer degree-level kinship relations between pairs of ancient individuals, including BREADR, correctKin, GRUPS-rs, KIN, and READv2. Additionally, many of these methods can distinguish between 1st-degree relation types (parent-child or full siblings).

However, the process for manually assembling these pairwise kinship relations into a pedigree can be very difficult and tedious. repare aims to automate this pedigree reconstruction process in a user-friendly, flexible manner.

11.1.2 Biological constraints

Certain biological information can aid in the reconstruction of ancient pedigrees. For example, mitochondrial haplogroups are transmitted through maternal lineages, and Y-chromosome haplogroups are transmitted through paternal lineages (for genetic males), so candidate pedigrees that violate these constraints can be discarded. Similarly, skeletal age-at-death estimates, runs-of-homozygosity (ROH) data, and temporal information can be used to constrain the set of candidate pedigrees.

11.1.3 Imperfect kinship inference

One factor complicating ancient pedigree reconstruction is the error rate of kinship inference. In manual pedigree reconstruction, one must routinely consider alternative degrees of relatedness for pairs of individuals; it is very rare that a kinship inference method can achieve 100% accuracy on an ancient DNA dataset. repare replicates this error-checking process by explicitly considering alternative degrees of relatedness for each input pair, unless otherwise specified by the user.

11.2 Getting started

11.2.1 Installation

We recommend installing repare using the provided conda environment file.

curl -O https://raw.githubusercontent.com/Narasimhan-Lab/repare/main/repare-environment.yml
conda env create -f repare-environment.yml
conda activate repare

11.2.2 Input files

repare takes two CSV files as input: a nodes file which describes individuals and a relations file which describes (inferred) pairwise kinship relations.

11.2.3 Nodes CSV

Each row represents one individual. The required and optional columns are:

Column	Required	Description
id	Yes	Individual identifier (cannot be fully numeric)
sex	Yes	Genetic sex: M or F
y_haplogroup	Yes	Y-chromosome haplogroup (* as wildcard; empty for females)
mt_haplogroup	Yes	Mitochondrial haplogroup (* as wildcard)
can_have_children	No	Whether the individual can have offspring (default: True)
can_be_inbred	No	Whether parents can be related at 3rd degree or closer (default: True)
years_before_present	No	Approximate birth date in years before present

Example:

nodes.csv

id,sex,y_haplogroup,mt_haplogroup,can_have_children,can_be_inbred,years_before_present
N1,F,,ab*,True,True,1000
N2,M,xyz,abc,True,False
N3,M,xyz,def,False,True,975
N4,M,*,g*,True,True
N5,F,,ghi,True,False
N6,F,,de*,True,True

11.2.4 Relations CSV

Each row represents an inferred kinship relation between two individuals.

Column	Required	Description
id1	Yes	ID of individual 1
id2	Yes	ID of individual 2
degree	Yes	Inferred kinship degree: 1, 2, or 3
constraints	No	Semicolon-delimited list of allowed relationship types
force_constraints	No	Whether the constraint must be enforced (default: False)

Example:

relations.csv

id1,id2,degree,constraints,force_constraints
N1,N2,1,parent-child,
N2,N4,1,parent-child;child-parent,
N2,N3,1,,
N2,N6,1,,
N4,N5,1,,
N3,N6,1,siblings,True
N1,N3,1,,
N1,N4,2,,
N3,N4,2,,
N4,N6,2,,
N1,N6,2,,

11.2.5 Constraints

The constraints column allows you to restrict the possible exact kinship relation types (within a certain kinship degree) for a given pair of individuals. This can be useful when you have additional archaeological or genetic evidence. The full list of allowed constraint strings can be printed with:

repare --print-allowed-constraints

The allowed values include:

• parent-child
• child-parent
• siblings
• maternal half-siblings
• paternal half-siblings
• maternal aunt/uncle-nephew/niece
• maternal nephew/niece-aunt/uncle
• paternal aunt/uncle-nephew/niece
• paternal nephew/niece-aunt/uncle
• maternal grandparent-grandchild
• maternal grandchild-grandparent
• paternal grandparent-grandchild
• paternal grandchild-grandparent
• double cousins

When force_constraints is set to True, repare can not override that constraint during reconstruction. When force_constraints is set to False, overriding a constraint counts as one “inconsistency” in the resulting pedigree.

11.3 Running repare

11.3.1 Basic command

repare -n nodes.csv -r relations.csv

11.3.2 Full command-line options

Flag	Description	Default
-n NODES	Path to nodes CSV (required)	–
-r RELATIONS	Path to relations CSV (required)	–
-o OUTPUT	Output directory	Current directory
-m MAX	Maximum candidate pedigrees to explore	1000
-e EPSILON	Epsilon parameter for greedy sampling	0.2
-s SEED	Random seed for reproducibility	42
-d	Disable pedigree plotting	Off
-w	Write outputs for alternate pedigrees with equivalent scores	Off
-v	Verbose output (INFO-level logging)	Off

11.3.3 Example

With the example files from above:

repare -n nodes.csv -r relations.csv -o outputs

11.4 Output

repare produces three output files:

11.4.1 corrected_input_relations.csv

This file includes the input kinship relations, as well as any corrections that repare applied. For instance, if a pair was originally inferred as 1st-degree, but repare determined it should be 2nd-degree to achieve a better pedigree, the corrected degree will appear here.

11.4.2 reconstructed_exact_relations.csv

This file contains the exact relation type for each pair (e.g., parent-child, siblings, paternal half-siblings).

11.4.3 reconstructed_pedigree.pdf

A plot of the reconstructed pedigree. This output requires the Graphviz/PyGraphviz dependencies from the conda installation.

Tip

Use the -w flag to output alternative pedigree reconstructions that have the same inconsistency score as the best one. This can be useful when the data does not uniquely determine a single pedigree.

11.5 Practical considerations

11.5.1 Choosing the number of candidate pedigrees

The -m parameter controls how many candidate pedigrees repare explores at each algorithm iteration. Higher values increase the chance of finding the optimal reconstruction but also increase runtime. The default of 1000 is suitable in most cases. You may wish to increase this parameter for larger pedigrees.

11.5.2 Haplogroup wildcards

When haplogroup information is uncertain, use the * wildcard. For example, R1b* is compatible with R1b1a or R1b1b. This can be useful when haplogroup calls are imprecise due to low coverage.

11.6 Test yourself

1. You have three individuals A, B, and C. A and B are 1st-degree relatives, B and C are 1st-degree relatives, and A and C are 2nd-degree relatives. A and B are male and share the same Y haplogroup; C is female. What pedigree structure(s) could explain these data?

2. An individual is female but has a Y-chromosome haplogroup listed in the nodes file. What problem would this cause for repare, and how should you handle it?

3. You run repare on a dataset, and it corrects 5 out of 6 input relations. What might this indicate about your data or your set of individuals?

4. When should you use force_constraints in the relations CSV?

11.7 Further reading

• repare GitHub repository: https://github.com/Narasimhan-Lab/repare
• Huang, Li, and Narasimhan (2026): the paper describing repare.
• Alaçamlı et al. (2024): the paper describing READv2, a widely used tool for estimating biological relatedness in ancient DNA.
• Rohrlach et al. (2023): BREADR, a Bayesian approach for estimating biological relatedness in low-coverage data.
• Lefeuvre, Marsolier, and Bon (2026): a benchmarking study comparing ancient kinship estimation methods.

Alaçamlı, Erkin, Thijessen Naidoo, Merve N. Güler, Ekin Sağlıcan, Şevval Aktürk, Igor Mapelli, Kıvılcım Başak Vural, Mehmet Somel, Helena Malmström, and Torsten Günther. 2024. “READv2: Advanced and User-Friendly Detection of Biological Relatedness in Archaeogenomics.” Genome Biology 25 (1): 216. https://doi.org/10.1186/s13059-024-03350-3.

Huang, Edward C., Kevin A. Li, and Vagheesh M. Narasimhan. 2026. “Fault-Tolerant Pedigree Reconstruction from Pairwise Kinship Relations.” Bioinformatics, May, btag251. https://doi.org/10.1093/bioinformatics/btag251.

Lefeuvre, Maël, Marie-Claude Marsolier, and Céline Bon. 2026. “Evaluating the Performance of Ancient DNA Genetic Relatedness Estimation Methods Using High-Fidelity Pedigree Simulations.” Genome Biology 27 (1): 132. https://doi.org/10.1186/s13059-026-04016-y.

Rohrlach, Adam B, Jonathan Tuke, Divyaratan Popli, and Wolfgang Haak. 2023. “BREADR: An r Package for the Bayesian Estimation of Genetic Relatedness from Low-Coverage Genotype Data.” bioRxiv. https://doi.org/10.1101/2023.04.17.537144.