How Accurate Is 23andMe?

Ancestry Composition Genetic Ancestry Predictions

The 23andMe Ancestry Composition Report provides an estimate of your ancestry breakdown by comparing your DNA to the DNA of more than 14,000 people from around the world to help you uncover clues about where your ancestors may have lived. With one of the largest reference datasets in the world, 23andMe provides customers with one of the most detailed and accurate ancestry breakdowns on the market.

23andMe’s Ancestry Composition analysis works by dividing your genome into thousands of segments and then estimating the ancestry of each segment based on which global population it most closely matches. The segment ancestries are then tallied up. Because Ancestry Composition breaks your genome into thousands of segments, our models can give us a view into very small portions of your genome (what we may call “highly precise”).

Our algorithms make ancestry estimates based on probabilities and they’re generally very accurate, but your results are not set in stone. We are always trying to improve and refine these estimates. Therefore, we’ve included ways for you to explore your results at 5 different levels of scientific certainty (i.e. “confidence”), and get more from them:

• With the Ancestry Composition Report, customers can view their ancestry percentages at 5 different confidence thresholds - 50% (default), 60%, 70%, 80%, and 90% (most conservative).
• For a given segment of DNA, 70% confidence in an assignment means there’s still a 30% chance the ancestry of that segment may be something slightly different: For example, instead of “Italian” ancestry (70% chance), maybe we instead predict “French & German” ancestry (30% chance) for that segment, because one of your ancestors once lived on the border between Italy and France and it’s hard to tell the difference.
• We choose to provide customers with ancestry composition percentages at the default confidence threshold of 50% so that they can see the “most likely” genetic ancestries found across their genome. If customers choose to see a more conservative estimate (where the model is more than 50% confident in each assignment), they can move the slider at the bottom of the Ancestry Composition Report by the Chromosome Painting tool.

There’s a little bit more that goes into how we initially train our system on a group of reference individuals (these individuals eventually become the reference population we use to assign your ancestry), and then how we test the system. You can read about it in our white paper, and see the results of these tests in our Ancestry Composition Guide (they are called precision and recall values). Our precision values for the vast majority of populations are greater than 90%, indicating high accuracy, meaning when we are confident in our ability to predict a piece of DNA as coming from one of our reference populations.

Similarly, the algorithm we use to calculate your Country Matches and Genetic Groups looks for pieces of DNA you share with individuals from around the world. There are more than 400,000 customers (who have consented separately to participate in 23andMe Research) in the reference populations for determining your Country Matches and Genetic Groups. Based on how much DNA you share, we assign a confidence to our prediction. We report these confidence levels in the Scientific Details section as “not detected,” “possible,” “likely,” or “highly likely.” “Possible” means we are between 30% - 49.9% confident in the assignment, while “highly likely” means we are at least 80% confident in the assignment. “Not detected” means we are less than 30% confident in assigning that Country Match to you, or less than 50% confident in assigning that Genetic Group to you.

DNA Relatives Relationship Predictions

The DNA Relatives feature is one of the most interactive features of 23andMe, allowing you to opt-in, find and connect with genetic relatives and learn more about your family. We’re extremely confident in the percentage of shared DNA we return to customers in this feature. However, the predicted relationship provided in this feature is just that -- a prediction based on the information we have. The reason we aren’t always 100% confident in the relationship prediction is that some relationships involve similar percentages of shared DNA. For example, it’s sometimes difficult for us to distinguish between a grandparent/grandchild relations and a half-siblings relationship. Because genetic information is hereditary, knowing something about your genetics also tells you something about those closely related to you. Click here for more information about what you might learn from 23andMe.

Our DNA Relatives feature is optional. 23andMe customers can opt-in or out of the feature at any time from within their account settings.

Learn more about the 23andMe Ancestry Service here

Our Genetic Health Risk* and Carrier Status* Reports meet FDA requirements for analytical and clinical validity.

• Analytical validity refers to the accuracy and reproducibility of the test -- our ability to accurately tell someone whether they have an A, T, C, or G at the particular genetic marker we're testing. Each variant in our Genetic Health Risk and Carrier Status Reports demonstrated >99% accuracy, and each variant also showed >99% reproducibility when tested under different laboratory conditions.

• Clinical validity refers to the strength of the association between a particular genetic variant and a particular health condition. In other words, how strong is the evidence that a particular genetic variant increases the risk of developing a specific health condition? To establish the clinical validity of a variant, we require strong evidence from the scientific literature, in the form of clinical guidelines or multiple peer-reviewed publications. To learn more about 23andMe's standards for clinical and scientific validity, please see our white paper.

Limitations

Even though our Genetic Health Risk and Carrier Status Reports meet FDA requirements for analytical and clinical validity, there are limitations specific to each report. For example, there can be thousands of variants associated with a specific health condition and our report may only look at a subset of those variants. This means if someone has one of the variants not tested, they could still be at increased genetic risk for the condition. 23andMe highlights the intended uses, limitations, and relevant ethnicities within each Genetic Health Risk and Carrier Status Report, and you can learn more about them by visiting 23andMe Test Info.

Reports Powered by 23andMe Research

In addition to our Genetic Health Risk and Carrier Status Reports, we have reports that are powered by 23andMe research. Reports and features that are "Powered by 23andMe Research" are developed by 23andMe scientists using data and insights gathered from thousands of customers who have consented to participate in our research.

• These types of reports are created and internally validated through extensive internal research and data analysis, rather than being based on external published research.
• These reports are reviewed by our team of scientific and medical experts to confirm that the information included is accurate and consistent with scientific knowledge.
• These reports have not been clinically validated. To learn more about the science behind one of these reports, please see our white paper on the Type 2 Diabetes report.

It's important to keep in mind that for most of the health conditions we report on, genetics is only one part of the story; a person's risk of developing the condition also depends on other factors like family history, lifestyle, and environment, which are not included in our test.

Learn more about our Health + Ancestry Service here

Our Wellness Reports explore how our DNA can influence lifestyle factors like diet, exercise, and sleep. Some of our Wellness Reports are based on specific genetic variants that have support from the scientific literature. Other Wellness Reports (the Genetic Weight report) are based on statistical models powered by 23andMe research. To learn more about 23andMe's requirements for clinical and scientific validity, please see our white paper.

Our Traits Reports explore how your DNA influences your appearance, your senses, and other characteristics, ranging from eye color to taste preferences to whether you can match a musical pitch. Although these reports predict your chances of having a trait based on your genetics, many other factors can influence your actual traits, including lifestyle, environment, and genetic variants not included in the report. That's why we present your results as a probability or a likelihood, not a certainty. For example, if 90% of 23andMe research participants with a certain genetic result have one version of a trait, that still means 10% of research participants have the other version of the trait.

Even though our prediction about your trait may not always match your actual trait, the genetic result underlying that prediction is highly accurate. Our genotyping platform is a well-established and reliable technology for analyzing DNA, and all samples are processed in CLIA-certified labs.

Learn more about our Traits Reports here