HumanCode is highly committed to using the most accurate and advanced available methods for translating your raw genetic data into useful insights.  Most often, these take the form of trait predictions.

Additionally, we are committed to providing as much transparency and educational material to help people understand the science.  We also make it a goal to never overstate the predictiveness of genetics.  As we like to say, "DNA is not destiny".  There is still much to learn about how our genes impact our minds and bodies, and the research gets better every day.

In BABYGlimpse, we go a step further in trying to give a glimpse into how the genetics of two people could be passed on (in DNAPassport we don't do any offspring trait prediction).  In some cases, we can predict how a simple trait could be inherited, using a method similar to the Punnet Square you may remember from high school biology.  In other cases, for complex traits, we simulate the combination of the couples' genetics, and present the most likely range of outcomes.

In all cases, we strive to present the information with probability indicators (e.g. percent likelihood), and also explain the known genetic vs. environmental influences on a given trait.

The reality is that there are a wide range of factors to consider when answering the question "how accurate is this prediction", and we want to give our customers as much information as they desire as they dig in to their own genetics.

Our trait prediction system relies on 2 main inputs:

  1. GWAS (genome-wide association study) mining: first, we research the best (based on number of participants, reputation of journal/author) studies we can find.  Sometimes we use aggregated databases of GWAS, like GWAS Catalog, to find them. We then extract "odds ratios" from the literature.  These ratios measure the impact that a given genetic marker might have on a trait for an individual, in comparison with the reference population for that individual.  Example: the genetic marker (or "SNP") of A (aka the "genotype"), at a certain location on a certain chromosome, means that individuals with that marker are 20% more likely to have the trait or condition studied (aka the "phenotype").
  2. In-House Predictive Models: Wherever possible, we use population datasets to develop our own scoring models.  This approach requires that we have access to datasets which correlate individual genetic data to the phenotypes we are looking at.  These datasets can be public or private.  The UK BioBank is one such resource.  We then use machine-learning algorithms, for example, neural networks, to develop models which can predict trait level outcomes based on a combination of genetic markers - often many thousands of them can influence a single trait.  Adult height is one such example - there are thousands of SNPs which correlate with adult height, and newer machine-learning models do a fantastic job of predicting height from genetics. 

The table below shows the 3rd party research that we are using, for each trait:

We understand that this information can be complex and intimidating - we're here to help!  We want our customers to understand as much as they can about genetics, inheritance, and how their DNA influences them.  

Please feel free to chat or email us (hello@humancode.com) for more information.

Did this answer your question?