By analyzing survey responses of 23andMe customers who are genetically similar to you, we can use predictive models to provide you with your likelihood of having the same trait. Most traits are influenced by many factors, both genetic and nongenetic, while others can be strongly predicted by the presence of just a few genetic variants. Depending on which trait we are looking at, we estimate your likelihood of having that trait using either a statistical model or a curated one.
Estimating Your Likelihood
One of the main differences between the two methods is how we choose which genetic variants to include in our analyses. The Statistical Model uses the 23andMe database to discover variants that are statistically associated with the trait, whereas the Curated Model uses variants from published scientific studies.
Most traits are influenced by many factors, both genetic and nongenetic, that we don't yet know very much about. For these kinds of traits, we use a statistical model to determine your likelihood of having or not having a trait. In the statistical model, we identify variants significantly correlated with a trait by using data from tens of thousands of 23andMe customers who have consented to research. Then, we use statistical methods to generate a "score" for that trait based on which variants you have, as well as your age, sex, and in some cases your ethnicity. We predict your likelihood of having different versions of the trait based on survey responses of 23andMe customers with similar scores. Read more about our statistical methodology.
Examples of reports based on the Statistical Model include: Cleft Chin, Cheek Dimples, Unibrow, Widow's Peak, Earlobe Type, Hair Color, Hair Curliness, Newborn Hair, Toe Length Ratio, Finger Length Ratio, Photic Sneeze, Freckles and Sweet Taste Preference.
Other traits can be strongly predicted by the presence of just a few genetic variants, as documented in published scientific studies. In the curated model, we take these well-studied variants and look at them in our own database. From the survey answers of customers who have consented to research, we calculate the percentage of customers who have the variant and also have the trait. From this, we estimate your likelihood of having the trait as well.
Examples of reports based on the Curated Model include: Eye Color, Red Hair, Earwax Type, Skin Pigmentation, Bitter Taste Perception, and Asparagus Odor Detection.
Predicted Trait vs Actual Trait
In some cases, the likely trait estimated by 23andMe will not match your actual trait. Your estimated trait reported by 23andMe is a likelihood, not a statement of certainty. It is the probability that you do or do not have a trait based on an analysis of 23andMe customers who are genetically similar to you at the markers analyzed for that report. Because the result is a probability, there is still a chance of not having the result we predict. For example, if we say "you are likely to have" a particular trait because 85% of customers who are genetically similar to you report having that trait, there is a still a 15% chance that you do not have that trait.
Your actual trait may not match the likely trait reported by 23andMe for a couple of reasons:
- Your particular version of the trait may be primarily determined by genetic variants not included in our analysis or by other factors altogether.
- Even when we find that 95% of customers with a certain genotype have one version of a trait, that still means that thousands of our customers with that genotype will fall in the other 5% and have the other version.
- You may have unique ancestry. When we don't have enough customers with certain ancestries, we aren't able to evaluate how well our trait predictions work in those groups. As more customers with diverse backgrounds participate in research, we'll be able to improve our likelihood models for everyone.
Although our prediction about your trait may not always match your actual physical trait, it's important to understand that the genotype 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 certified labs.