Ancestry is a company providing Geneology and Direct-to-Consumer (DTC) Autosomal Genetic Testing services.
Ancestry offers two major genealogical services based on a saliva test: identity-by-descent analysis (see community detection) and genetic ancestry. ethnicity
- non-sex chromosomes.
- Like 23andMe, AncestryDNA uses SNP-chip technology, which probes for 730,525 genomic variants. ethnicity
- Specifically, they use the Illumina OmniExpress Chip. han2017clustering
- The genomic variants chosen for the array were selected based on the Human HapMap Consortium, an international effort to catalogue global SNP variation. ethnicity
- The array only detects SNPs with an allele frequency > 5% ethnicity
- The majority of the SNPs are designed to target European populations.
- Until recently, AncestryDNA tests only provided an analysis of an individual’s distant ancestry. curtis2017estimation
- This is achieved by comparing SNPs across the individual’s genome with those characteristic of various ethnic groups.
- Each ethinic group has characteristic SNPs as result of mating barriers between these ethnic groups.
- Ancesty uses a program known as ADMIXTURE which estimates the proportion of an individual’s membership in a set of ancestral populations based on his or her genotype. ethnicity
Now, Ancestry is working on a service to identify an individual’s more recent ancestral groups using identity-by-descent inference. han2017clustering
Ancestry’s large dataset permits them to identify individuals who share haplotypes from a common ancestor. han2017clustering
- The probability that a particular region in the genome is shared between two descendants of a common ancestor four generations ago is <1%.
- However, Ancestry has collected over 500 million such pairs.
Recent ancestry is analyzed using haplotype variation. curtis2017estimation
Haplotype: group of alleles inherited by a single ancestor.
The familial relationship between two individuals can be determined using the length of shared DNA between them.
- e.g. a parent and a child will share half of the genome.
This is achieved in two steps:
A genetic network of 742,394 consenting individuals was created. An unsupervised Machine Learning community detection i.e. clustering algorithm was used to identify haplotypes characteristic of various genetic groups.
- The Louvain method was used.
- Community detection was performed recursively to discover sub-communities.
The communities discovered through clustering of the 742,394 individuals are used as training data for unsupervised classification algorithms. One algorithm is used for each community so that an individual may be assigned to multiple communities.
Challenges of community detection curtis2017estimation
- Clustering is computationally intensive and is not feasible to perform regularly.
- Clustering algorithms typically assign nodes to only the single most suitable cluster.
- To solve this, Ancestry built a classification algorithm for each community.
- While clustering is consistent across an entire database, individual assignments may change across repetitions.
Ancesty offers identity-by-descent genotyping with the goal of allowing customer’s to reunite with lost relatives.
However, this very promise has the potential for exploitation by law-enforcement agencies—nearly 60% of searches for individuals of European descent will result in a third-cousin or closer match. identity
Notably, in a recent case, law enforcement used a genetic search to trace the Golden State Killer—authorities were able to locate the killer’s third cousin.
- Although this particular case is positive, the same technique could be used for nefarious purposes, such as re-identifying subjects of research studies from their genetic data.
- Currently, the U.S department of Health and Human Services does not consider genome-wide datasets as identifiable information.
DTC genetic testing companies provide customers with their raw DNA files which can then be uploaded to other services.
[ethnicity] Noto, Wang, Song, Turissini, Sedghifar, Garrigan, Starr, Byrnes, Hong, Ball & others, Ethnicity Estimate 2018 White Paper, , . ↩
[han2017clustering] Han, Carbonetto, Curtis, Wang, Granka, Byrnes, Noto, Kermany, Myres, Barber & others, Clustering of 770,000 genomes reveals post-colonial population structure of North America, Nature communications, 8, 14238 (2017). ↩
[curtis2017estimation] Curtis & Girshick, Estimation of Recent Ancestral Origins of Individuals on a Large Scale, 1417-1425, in in: Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, edited by (2017) ↩