Researchers at The Hospital for Sick Children have developed a new machine learning model to improve the accuracy of prenatal genetic testing. This advancement addresses a major challenge in clinical genetics, where nearly one-third of detected genetic changes are labeled as variants of uncertain significance. These variants create ambiguity for doctors and families, as it remains unclear if the genetic change is harmful or harmless.
The team, led by Rosanna Weksberg and Sanaa Choufani, focused their efforts on epigenetics and episignatures. These chemical tags in DNA indicate specific genetic conditions. Previously, these signatures were tissue-specific, meaning patterns found in blood could not be used to analyze other samples like amniotic fluid or placental tissue. By creating a tissue-agnostic model, the researchers have removed this barrier to clinical application.
In a study published in The American Journal of Human Genetics, the team demonstrated the capability of their model by testing it against data from individuals with and without Down syndrome. They trained the system on DNA methylation data covering six different prenatal and postnatal tissue types. The model successfully identified the Down syndrome pattern in every instance, proving that blood-derived signatures can reliably identify markers across different types of human tissue.
This technology points toward a future where diagnostic testing is faster and more accessible. By enabling the use of less invasive samples like saliva or oral swabs, the method could eventually reduce the reliance on blood draws or other invasive collection procedures. The objective is to provide families with clear answers sooner, shortening the time spent searching for a diagnosis for rare disorders. This work marks a move toward higher precision in prenatal care and provides a framework for investigating conditions where traditional tissue sampling is difficult or impossible.