In contrast, the recall of Ancestry Composition for these tests was only 0.95 or less, and the recall ofall our models was less than 0.9 in every test. It is clear that Ancestry Compositionis much more accurate at predicting ancestry than our current in-house population-genetics models. For our currentmtDNA and Y-chromosome tests, we have much lower recall, which is not surprising given the known problemsthat we have had in assembling the reference database. However, our models perform much better on these tests than onthe autosomal tests, which we are analyzing independently.
When we looked at the 21 autosomal tests, we found that Ancestry Composition was more accurate at predictingancestry than our in-house population-genetics models in 20 out of 21 tests, and these tests used a minimum confidencethreshold of 50%. The recall of Ancestry Composition was greater than 0.99 in 21 of the tests, and the recall ofall our models was only 0.9 in two tests.
Since Ancestry Composition is an ensemble of models, we calculated both the precision andrecall of Ancestry Composition by repeatedly sampling from the 20 percent of individuals in the testset held out during the quality-control test. Precision is independent of recall, so bothmeasurements are important.
For a given query, we will sample the target population 100 times and report theprecision and recall as the average over the sampled population. To calculate precision, we divide thenumber of correct predictions by the total number of predictions. To calculate recall, we divide thenumber of correct predictions by the total number of actual population members.
The MATLAB environment is based on the MathWorks MATLAB® Compiler, or MMC, which is a proprietary language developed by MathWorks to facilitate the building of applications. MMC understands the MATLAB language and various task APIs and provides efficient intermediate language compilers for compiling functions and scripts.
The MATLAB® Engine is the core of the MATLAB product. It compiles, executes, and interprets MATLAB scripts and programs, as well as runs the M-files and M-links created by the MMC. It also allows the exchange of data between MATLAB applications, the MathWorks M-files and M-links, and other programming languages, including C, C++, and Visual Basic. 827ec27edc