Local Ancestry Determination
Local ancestry inference (LAI) is an indispensable part of any genomics research project. A large proportion of individuals exhibit some degree of admixture and genetic diversity at the sub-continental level is increasingly relevant from a medical point of view, even at the finest scales. Understanding the criticality of this level of granularity, we have developed Orchestra, our ancestral deconvolution tool.
As seen in Figure 1, Orchestra markedly outperforms other leading ancestry methods in both non-admixed (generation 0) and admixed samples (generations 1-6). It has ~15% better overall recall and ~14% better overall precision than the next-best model. It also retains high accuracy across all tested populations, with a remarkable ability to distinguish between closely related ancestries. Orchestra achieves an accuracy of over 75% for 100% of populations within the 1KGP dataset and for 75% of populations within the custom data panel. The other models struggle with about a third of the populations, where their accuracy falls below 50%.
In Figure 2, you can see that Orchestra not only outperformed the other state-of-the-art models in highly mixed samples such as Latinos but was also able to collect the genetic history of Latin Americans with greater precision than comparable algorithms..