2. Genotype phasing algorithm
2.1. Introduction
As explained in section 1.2, genotypes alone often cannot tell us which allele copy was inherited from the father and which was inherited from the mother. One exception to this is when we have genotypes sampled from both parents and child (called a trio). In this case, since the laws of genetic inheritance tell us that alleles can only be transmitted from parent to child in specific ways, we can use this information to very accurately assign alleles to each of the two chromosome copies. However, since we cannot depend on all customers taking the AncestryDNA test with both their parents, we need a more sophisticated approach that can accurately determine the phase of the genotypes—the assignment of alleles to chromosome copies—without parental information.
The strategy is to simultaneously phase the genotypes from a large number of unrelated individuals. Since the genotypes observed at consecutive SNPs can be phased in many different ways, the basic principle is to prefer a phase that results in two sequences on each of the chromosomes that are also observed in many other samples. In other words, if the phasing yields a sequence that is unique, this is probably the wrong way to phase the genotypes. This principle is based on the expectation that short sequences, called haplotypes, are typically shared by many people in a large population.
Practically speaking, this means that we have a chicken-and-egg problem: accurate phasing of the genotypes requires us to determine the more common sequences (the haplotypes); to determine the more common haplotypes, we need to first know how to phase the genotypes. Fortunately, software such as BEAGLE (Browning and Browning, 2007) and HAPI-UR (Williams et al., 2012) are designed specifically to solve this chicken-and-egg problem from thousands of samples simultaneously.
An important benefit of such approaches is that phasing accuracy improves as more unrelated samples are analyzed simultaneously. Therefore, in principle we could achieve highly accurate phasing by simultaneously analyzing the hundreds of thousands of genotype samples from AncestryDNA customers. However, as fast as methods like BEAGLE and HAPI-UR are, they were not designed to jointly handle millions of samples.
Therefore, we have developed a modified strategy, which we call “Underdog.” Underdog learns haplotype frequencies—or, more precisely, frequencies of “haplotype clusters”—in a large number of AncestryDNA samples. Then, once we have learned haplotype cluster frequencies, we use Underdog to quickly phase the genotypes of new customers.
Originally Posted by Cumansky
Reply With Quote
Bookmarks