Modeling offspring with a partner

The Lab takes your genome and a partner's genome and projects what an offspring of the two of you would inherit. The model has two main layers: a Mendelian layer that handles single-gene traits exactly, and a polygenic layer that handles complex traits probabilistically. Both layers run for every projected embryo, so the resulting profile contains hundreds of traits per child.

The Mendelian layer

For autosomal recessive conditions, an offspring is at risk of being affected only when both parents carry a pathogenic variant in the same gene. When only one parent carries, every child has a 50% chance of being a carrier and a 0% chance of being affected. The Lab walks through every recessive condition in the screening panel (89 genes today, expanding) and computes the per-trait inheritance probabilities exactly.

For autosomal dominant conditions, a single copy from either parent is sufficient. Each affected parent passes the variant to half of their offspring on average. For X-linked conditions, the inheritance pattern depends on the sex of the child, and the Lab computes that conditional probability separately for XY and XX projections.

The polygenic layer

For complex traits like height, BMI, type 2 diabetes risk, schizophrenia risk, or cognitive function, the Lab computes a projected polygenic score for each embryo by simulating the meiotic shuffle. Each parental chromosome is broken into segments of average size 50-75 megabases (matching the published recombination rate), and each segment is independently inherited from one of the two parental haplotypes. The resulting child genome is fed into the same polygenic scoring methods used for adults.

Because the meiotic shuffle is stochastic, the projected polygenic score for any given trait varies from embryo to embryo. The platform reports the per-embryo z-score along with a 95% confidence interval that reflects both the meiotic shuffle and the residual error of the underlying PRS model.

What the Lab cannot do

• It cannot guarantee any phenotype. A polygenic prediction with R² of 22% leaves 78% of the variance unexplained, and the realised height, intelligence, or risk for any specific child can sit anywhere within a wide interval around the projected value.
• It cannot model de novo mutations, which arise during the formation of the egg and sperm and are present in roughly 60-100 per child. De novo mutations are rare per gene but real in aggregate, and they account for a meaningful fraction of severe paediatric disease.
• It cannot predict environmental contributions, gene-environment interactions, or rare variants outside the screening panel.
• It cannot replace pre-implantation genetic testing on actual embryos. The Lab projects what could happen across many possible offspring; PGT-A and PGT-M analyse a real biopsied embryo.